…protocols

Change text mentioning the "fdo" platform plug-in to "wl", which was
missing from Igalia#351.

Arrange to call drmFreeDevices() inside check_drm() and init_drm()
to avoid leaking the arrays of devices returned by drmGetDevices2().

Pass the correct pkg-config module name to gi-docgen depending on the
version of libsoup in use: WPE WebKit builds with libsoup2 have API
version 1.0, and builds with libsoup3 have API version 1.1.

This avoids not finding any environment setup script due to the CPU name
and features changing when SDKs are updated: use a wildcard to make sure
at least one gets picked.

Use GObjectClass.finalize instead of CogPlatformClass.teardown. Now that
CogPlatform derives from GObject, it is possible to reuse its functionality
instead of having an ad-hoc vfunc that gets called immediately before
destroying an instance. As a bonus, not only the implementations follow
a more idiomatic GObject approach, but also it is not possible anymore
to forget a call to cog_platform_teardown() before destruction.

Update the check-style script and the CI job to use clang-format 13, which
is the first version that supports combining The "PointerAlignment: Right"
with "AlignConsecutiveDeclarations: true". This was unimplemented until
recently, see https://reviews.llvm.org/D103245

While at it, do not let clang-format choose a different way of doing
pointer alignment by setting "DerivePointerAlignment: false".

Retry the curl command up to ten times in case of download failures.
This should alleviate build issues like the following, where the server
sometimes seems to wrongly close HTTP/2 streams, or when some connection
issue results in partial downloads:

    https://github.com/Igalia/cog/runs/4176134469
    https://github.com/Igalia/cog/runs/4176255670

While at it, pass "-C -" to let curl determine from which point to retry
the download. As the toolchain packages are big, this should help avoid
timeouts.

Finally, use HTTP/1.1; because using HTTP/2 seems to be a source of
sporadic issues, and for a big, bulky download, the gains from HTTP/2
(like multiple streams and faster connection open) are basically none
due to the transfer itself taking most of the time.

…protocols