[ChandraChaan]

AudioFifo.write() checks that every later frame has the same sample format, sample rate and time base as the first one, but not the same channel layout. The FIFO is allocated with the first frame's channel count, so:

• writing a stereo s16 frame into a FIFO that started with mono reads only half of the frame's data and scrambles the samples, with no error;
• writing a frame with fewer channels than the first reads past the end of the frame's buffers (for planar formats, data[1] is NULL).

The channel layout is now compared with av_channel_layout_compare, like AudioResampler.resample already rejects frames whose layout differs from its template. The template keeps the first frame's layout via _copy_internal_attributes, so the comparison is against what the FIFO was allocated for.

Added test_layout_mismatch to tests/test_audiofifo.py.

I ran the smoke workflow on my fork:

• Test only (current master): test_layout_mismatch fails with Failed: DID NOT RAISE ValueError (1 failed, 600 passed) on the Ubuntu and Windows jobs. https://github.com/ChandraChaan/PyAV/actions/runs/41539175134
• With the fix: Windows, three Ubuntu jobs (FFmpeg 8.1.3 and 9.0.2) and the armv7 smoke pass. https://github.com/ChandraChaan/PyAV/actions/runs/41539180238
• make lint passes: https://github.com/ChandraChaan/PyAV/actions/runs/41539186508

The macOS jobs were cancelled on my fork twice (with and without the fix) before running, and the RISC-V job never got a runner, so I have no macOS or RISC-V result.