fix(*): elminate flickering in interpolated output and improve color consistency with input

[exwm]

Related Issues

• fixes Flickering problem on the video screen after using RIFE frame interpolation #1336
• closes hqdn3d ffmpeg filter (bootleg solution to eliminate flicker from RIFE) #1364

Change Details

Add a virtual Interpolator::normalize() method that runs pass-through frames
through the same pixel-format round-trip as interpolated frames, eliminating
temporal colour inconsistencies and flickering in alternating frames (assuming 2x interpolation).

Refactor in_time_base_, out_time_base_, and out_pix_fmt_ from InterpolatorRIFE private members into Interpolator protected members to avoid duplication in future derived classes.

Capture authoritative color properties (color_space, color_range, color_primaries, color_trc) from the decoder context once during init() and store them in the Processor base class. Introduce make_color_hint(), which creates a metadata-only AVFrame carrying these properties and forwards it to avframe_to_ncnn_mat() and ncnn_mat_to_avframe(), ensuring sws_scale always uses a consistent colour matrix instead of falling back to BT.601 defaults whenever a frame carries AVCOL_SPC_UNSPECIFIED.

Additional changes in src/conversions.cpp:

• Add avcol_spc_to_sws_cs() to map AVColorSpace to SWS colorspace constants with a BT.601 fallback and a thread-safe one-shot warning via std::once_flag.
• Replace hardcoded SWS_BILINEAR with static constexpr int SWS_FILTER = SWS_BICUBIC.
• Extract apply_sws_colorspace() helper to deduplicate the two sws_setColorspaceDetails call sites; failures emit a one-time warn on the first occurrence followed by debug logs with a running count thereafter.
• Validate the return value of sws_scale() against dst_frame->height before returning.

Register src/interpolator.cpp in CMakeLists.txt.

Testing

• Interpolated test video with options --frame-rate-mul 2 --processor rife --rife-model rife-v4.26 --pix-fmt yuv420p -c libx264 before and after this change -> observed flickering in video due to alternating artifacting/noise patterns eliminated
• Upscaled video with options --processor realesrgan --scaling-factor 2 --realesrgan-model realesr-animevideov3 --pix-fmt yuv420p -c libx264 and output before and after the change seems the same, given that the input video used bt709 and yuv420p which may not exercise some of the new defensive code around color space handling.