Implement support for proper HDR to SDR conversion

[k4yt3x]

This ticket tracks the proper support for processing HDR videos.

When processing HDR videos, the decoded frames could have properties looking like this:

• Pixel Format: yuv420p10le
• Color Primaries: BT.2020
• Color Transfer: SMPTE2084 (PQ)
• Color Range: TV (limited range)

Before the frame can be passed to, say, Real-ESRGAN, it must be converted into an ncnn::Mat, which will require converting the 10-bit HDR image into an 8-bit SDR (RGB24) image. Right now, AVFrame conversions are done through libswscale, which can convert the frame formats, but does not support proper tone-mapping from HDR to SDR. As a result, processing HDR videos will result in color inaccuracies in the output video.

Currently, the idea is to use libavformat's zscale and tonemap filters to perform the tone-mapping and format conversion:

zscale=transferin=smpte2084:primariesin=bt2020:transfer=bt709:primaries=bt709:rangein=tv:range=tv,tonemap=hable,format=rgb24

Please feel free to suggest better ways to implement this feature and discuss about this in this thread if you're familiar with HDR content processing best practices.