vpx 10-bit colour encoding support

[totaam]

Follow up from #909, we may want to wait for vp10 to come out too.

The API documentation is very limited but both gstreamer and ffmpeg have support for it - which we could re-use via the enc_ffmpeg encoder (#1107).

[totaam]

libvpx needs to be compiled with --enable-vp9-highbitdepth, CONFIG_VP9_HIGHBITDEPTH should be set and the image formats we can use are "Image uses 16bit framebuffer":

VPX_IMG_FMT_I42016    = VPX_IMG_FMT_I420 | VPX_IMG_FMT_HIGHBITDEPTH
VPX_IMG_FMT_I42216    = VPX_IMG_FMT_I422 | VPX_IMG_FMT_HIGHBITDEPTH
VPX_IMG_FMT_I44416    = VPX_IMG_FMT_I444 | VPX_IMG_FMT_HIGHBITDEPTH
VPX_IMG_FMT_I44016    = VPX_IMG_FMT_I440 | VPX_IMG_FMT_HIGHBITDEPTH

I believe those formats store the the YUV channels in the high 10/12 bits of each 16bit value. Wasting quite a bit of space and requiring a CSC step to go from R210 to this format.. sigh.

Note: vpx now also supports alpha channel: VPX_IMG_FMT_HAS_ALPHA.

Also worth considering colorspace bits (#1155): vpx_color_space.

[totaam]

Somewhat blocked by rpath issues: #1405

[totaam]

re-scheduling - the 16-bit per channel reformatting constraint is costly.

[totaam]

Done in r26980.

Found a new scaling related bug: #2845.

For further reference, see ImageDepth.

[totaam]

Caused a regression with the vp9 encoder with older versions of libvpx (ie: Debian Buster and Ubuntu Focal): #3082