fix: Replace get_nowait() with get(timeout=5) in multiprocessing test

[Lorquas]

Problem

TestIni.test_queue in tests/test_multiprocessing.py is flaky. The child process calls _q.get_nowait(), which raises _queue.Empty if the serialized INIConfig object hasn't arrived in the queue yet. This is a race condition between Process.start() and the queue's internal pipe/buffer.

The failure is rare on a normal workstation but triggers consistently (~20% failure rate) in resource-constrained or containerized build environments — specifically Koji mock builds using systemd-nspawn, which is how CentOS Stream / RHEL RPMs are built. The %check phase runs runtests.py, hits this race, and fails the entire package build.

ERROR: test_queue (tests.test_multiprocessing.TestIni.test_queue)
  File "tests/test_multiprocessing.py", line 62, in getxy
    _cfg = _q.get_nowait()
_queue.Empty

Fix

One-line change: replace _q.get_nowait() with _q.get(timeout=5) in the child process. This gives the forked process time to receive the serialized object through the queue, matching the pattern already used by the parent on the result queue (w.get(timeout=1)).

Reproducer

Stress test with 200 parallel runs under CPU contention:

# Saturate all cores to simulate Koji builder pressure
stress-ng --cpu $(nproc) --timeout 120s &

# Run 200 parallel test invocations
for i in $(seq 1 200); do
  python3 -m pytest tests/test_multiprocessing.py -x -q > /tmp/mp_test_$i.out 2>&1 &
done
wait

grep -l "FAILED" /tmp/mp_test_*.out | wc -l

	Before	After
Failures (out of 200)	39 (20%)	0

[cnsnyder]

LGTM