This is the very first step in resolving issue #507. It moves the text for
the documentation of a single command out of the source code and into a
standalone text file. It sets up the framework for converting such files
back into C source for use by the optget() function.

Related #507

This builds on the prior change to extricate builtin command documentation
out of the source and into an external text file for each command.
Which is the next step into converting the documentation to a modern
markup language.

Related #507

This is to facilitate extracting the command documentation into text
files separate from the source code.

Related #507

Move the libcmd builtin documentation into files separate from the
source code.

Related #507

Related #507

Move the code for `b_readonly()` into its own module. Refactor it so
that the `readonly` and `export` commands only share the `setall()`
function.

This also moves `b_unalias()` into the same module. TBD is refactoring
the code to move the option parsing out of the `unall()` function and
into the respective `b_*()` functions.

Builds on Copr and OBS were failing because `print_scan()` function
definition was not prefixed with `static_fn`.

Recently builtins were moved into separate modules. It broke builds in
Copr and OBS. `print_scan()` function should be declared non-static as
it's used in more than one file. Fix made in previous commit was
incorrect.

Move each of the `bg`, `fg`, and `disown` implementations into separate
modules.

Move each command implementation into its own module.

Separate the break/continue and return/exit implementations into separate
modules to improve clarity. This increases the size of the binary by a
few words but that is well worth it for the improved clarity.

Separate the `cd` and `pwd` implementations into separate modules to
improve code clarity.

Fixes #1387

Resolves #1351

With this change all special builtins are now in their own source
module. More importantly this eliminates a lot of cleverness that
obfuscates the actual behavior.

Update fallback version to `2020.0.0-beta1`

We want to be able to separate `b_trap()` and `b_kill()` into separate
source modules. This is a step towards doing so.

The `Bltin()` macro serves no useful purpose. Its use does not affect
the behavior of the target functions. Which means those functions can
recognize long name options. Similarly command definitions which use
`bltin()` aren't required to use long options.

Resolves #1388

Fixes #1390

Related #1391

The e_astbin var doesn't serve a useful purpose so replace uses of it
with the string constant assigned to the var.

This var will be useful for doing things like finding the man pages
for builtins. It will hold the value of `meson --prefix=/path`.
Typically, /usr or /usr/local.

The use of `#define` with hard-coded offsets into a large table is
fragile. That is, dangerous because the offsets may not be correct.
Replace those with vars that are updated during process initialization
to point to the correct data structure.

Fixes #1394

The code that checks if the end of a `struct shtable2[]` has been reached
presumes a valid, non-NULL, pointer to the `sh_name` structure member.
It is more idiomatic to simply check for sh_name == NULL and change the
table initialization accordingly.

Having an independent set of vars contain pointers also stored in
shgd->bltin_nodes is sort of gross. Replace that approach with a
structure. That allow us to use a name instead of a possibly incorrect
array offset for quick access to a specific shell var. Now we only have
to worry about the number of vars changing and that is something we can
cheaply detect at run time.

Also implement a consistent naming scheme for the special symbol for each
shell var. The old scheme was very inconsistent. The new scheme has a
`VAR_` prefix for all of the vars and the suffix is the actual shell
var name.

Fixes #1398

This is a proof of concept change to implement the documentation using
Sphinx/RST markup. It includes documentation for the `alias` special
builtin and the `basename` command builtin.  The sole purpose is to
establish that using Sphinx/RST is viable by providing concrete examples
for examination.

I've deliberately chosen to include the documentation build artifacts
in the commit.  This means that for now the CI builds, and anyone else
building from source, doesn't need to have Sphinx installed to build the
documentation. This does mean that some care is needed to always run `make
man` and `make html` before committing any change to the documentation. TBD
is whether a trigger of some sort will be needed to ensure the two sets
of files are kept in sync.

Related #507

Another step in switching from the DocOpt base option processing done by
the AST `optget()` function to the borg standard `getopt_long()`.

It also introduces a system-wide ksh config script to enable auto-loading
functions included with ksh by default. This allows the use of the
auto-loaded `_ksh_print_help` function. As well as the dirs/popd/pushd
set of functions that have been the sole functions included with ksh
forever but not automatically enabled.

Related #507
Fixes #835

This adds a `man` function to give priority to ksh man pages. Making it
possible to do `man alias`, for example.

Related #507

This includes changes to make the b_alias, b_basename, and builtins unit
tests pass in light of the previous changes.

Note that a lot more work is needed. The logic shared by b_echo(),
b_print(), and b_printf() needs to be refactored out of b_print() and
into its own function; sh_print()?

instead of `MESON_INSTALL_PREFIX`.

This variable prefixes `DESTDIR` if set. This variable points to
temporary installation directory used during package builds. This fixes
package builds on OBS and Copr.

Recent changes have introduced new files that are installed by default.
Add those files under `%files` section in rpm spec files.

Unsurprisingly the AST `optget()` function defaults to POSIXLY_CORRECT
option parsing behavior. So switch the `getopt_long()` flag specs to
that behavior be prepending a `+` to the short flags string.

Use getpwuid to find the user name and home directory. Cache the user name for
future logins_tree lookups to avoid repeating the getpwuid call.

Fixes: #1391

The existing code is fine for long flags and single short flags. But
when short flags are bundled together if one of them is invalid then the
wrong arg is displayed in the error message.

The `getopt_long()` symbol `no_argument` is zero on every platform I have
access to. Nonetheless, don't assume that is true. Use the symbolic name.

If the sphinx-build command is installed use it to generate the
documentation from the reStructuredText source. Otherwise, use the
prebuild man pages.

Fixes #1402

This reverts commit 695269a.

The b_hist() code is structured weird. Switching to getopt_long() is
going to require a better understanding of the current structure.
Specifically, how an argument like `-1` is handled.

Fixes #1403

meson test outputs "SKIP" in the log when it sees this status.

https://mesonbuild.com/Unit-tests.html#skipped-tests-and-hard-errors

postscript.sh: cap error count to 50

meson uses specific high-numbered exit status values to detect skips and
errors. Avoid using these for legitimate test failures.

While investigating a security issue it came to light that the jobs
table isn't initialized early enough. Prior to this change command
substitution might happen before it is initialized. Which leads to
dereferencing a NULL pointer.

This was prompted by the need to correctly handle negative integers
which would otherwise look like invalid short flags. But it also fixes
the information displayed when an invalid flag is seen. The previous code
worked correctly only in limited situations (e.g., the invalid flag being
the first CLI arg).

The `X` short flag was left over from experimenting with the behavior of
`getopt_long()`.

To have `head` correctly handle a line number that looks like a flag
we have to use `getopt_long_only()` rather than `getopt_long()`. Add
a regression test to ensure this bug doesn't occur again.

Despite the embedded documentation implying negative ints are supported
they aren't. Furthermore, using a process group ID in the context of `bg`
and `fg` doesn't really make any sense.