Fix decoding of HEVC streams with intra-refresh and current Git FFmpeg
code. When intra-refresh is enabled and we are not catching the first
frame, no frame will be decoded without the flag
AV_CODEC_FLAG_OUTPUT_CORRUPT. This corresponds with FFmpeg commit
bc1a3bfd (2025-02-20), not yet included in current 7.1.1 / lavc 61,
should appear in lavc 62.
Steps to reproduce:
```
uv -t testcard -c lavc:e=libx265 -d gl
```
the behavior is now very slightly modified but effectively the same,
eg. change 1 to 3 channels copies now channel 0 to 1 and then 1 to 2
(before it copies 0 to both 1 and 2).
- do not exit if font failed to set (maybe the system one will be tried?)
- removed implicit helvetica (seem to be set by IM6 by default, anyways
so just not setting it won't solve the problem when helvetica not found
in sys)
If font is not set explicitly, MagickQueryFontMetrics fails with:
```
[text vo_pp.] MagickQueryFontMetrics failed: unable to read font `helvetica' @ error/annotate.c/RenderFreetype/1658!
```
(using ImageMagick 7.1.1)
- suppress the output for unit_evaluate
- improve debug printout codec_conversion_test_testcard_uyvy_to_i420
- improve C variant of ASSERT_EQUAL_MESSAGE macro (prints num values
also in hex)
- remove VLA from codec_conversions_test.cpp (not valid in C++)
fix test codec_conversion_test_testcard_uyvy_to_i420 fail caused by the
merge of testcard_common toI420 to to_lavc uyvy_to_yuv420p
Last line was incorrectly checked in case of odd number of lines.
Workaround for SDL3 mixer, that doesn't respect selected ch_count=1 now.
Reduced the channel count later in audio stack (this will work for other
acaps as well).
On mac, we can link either to MoltenVK or the vulkan-loader. The MoltenVK
variant is default, because the bundler then bundles the library as the
compat layer. If linked to vulkan-loader, Vulkan will likely not work if
the MoltenVK is not installed on the host computer.
But it is useful for debugging (the behavior is slightly different if
loaded via vulkan-loader even if the backend is MoltenVK - the KHR
portability bit/enumeration stuff).
1. to SDL3 nv12 (not supported before)
2. to_lavc - use rgba_to_bgra - the vc_copyline version was not keeping alpha
(if ever used - this may cause regressions, unfortunately, if alpha is
assumed to be 100%, because the orig version set it always)
3. uyvy_to_i420 use by testcard and to_lavc
uyvy_to_i420: do not write out of bounds
If width % 2 = 1, 1 extra luma on first line of pair will override the
first luma on the second.
