Passing logcfg to Step.from_cmdline (and likly other config uses) changes global state

[braingram]

There are a few tests in jwst that use logcfg:
https://github.com/spacetelescope/jwst/blob/8efc26fdbda7936199d1a05cb57707d2c1cb0665/jwst/pipeline/tests/test_calwebb_image2.py#L82-L84
https://github.com/spacetelescope/jwst/blob/8efc26fdbda7936199d1a05cb57707d2c1cb0665/jwst/pipeline/tests/test_calwebb_image2.py#L82C5-L82C11

These change global logging state and can lead to later tests that check log output failing. Here's a somewhat minimal example:

import jwst
from stpipe.config_parser import logger as log
from jwst.stpipe import Step

LOGFILE="foo.log"
LOGCFG="log.cfg"

logcfg_content = f"[*] \n \
    level = INFO \n \
    handler = file:{LOGFILE}"
with open(LOGCFG, 'w') as f:
    f.write(logcfg_content)

args = ["dq_init", "some_uncal.fits", f"--logcfg={LOGCFG}"]
print(log.level)
Step.from_cmdline(args)
print(log.level)

print("Log contents")
with open(LOGFILE, "r") as f:
    print(f.read())

print("Second run, no config")
args = ["dq_init", "some_uncal.fits"]
Step.from_cmdline(args)

print("Log contents")
with open(LOGFILE, "r") as f:
    print(f.read())

Running the example the first call to from_cmdline runs as expected logging messages to foo.log. However the second run (when no logcfg is provided also logs to the same file (appending log messages) and runs at the same level as configured in the first run.