Follow-ups found while fixing the audit's high findings (#469) #470

Found while fixing #469; each was reproduced or is a direct sibling of a fixed defect, and none is fixed there.

• A tool that changes the frame rate keeps the source's start timecode at the old rate. proxy.py --fps 50 and denoise.py --fps 50 on a 25 fps file starting 01:00:00:12 write 01:00:00:12 into the 50 fps output (the copied tmcd track's label). fix: the audit's high findings (captions, timing, audio, colour, batch), MCP argument safety, and Scorecard hardening #469 fixed fit (relabel + relabel_timecode= in run_keeping_subtitles) and cut. Not yet checked: background, caption, crop, grid, insert, join, multicam, redact, sequence, sphere, straighten, waveform.
• MXF: cut.seek_keyframe reads the wrong keyframe. For -ss 1.5 it reads 0.96 s while FFmpeg lands on 1.44 s, so an MXF stream copy writes no timecode (listed as dropped) and start_snapped is probably wrong for MXF too.
• metadata.py fails on a timecoded MOV into .mp4 (kind: ffmpeg, "Could not write header"): its -map 0 -c copy carries the tmcd data track.
• FFmpeg 8.0 converts a tagged/untagged mix in other mixing filters. fix: the audit's high findings (captions, timing, audio, colour, batch), MCP argument safety, and Scorecard hardening #469 labels an untagged PiP beside a tagged main in overlay.py --video (8.0.1 moved the main from 80/110/192 to 88/103/192). broll, grid, join and multicam label only beside a converted input; whether 8.0 shifts a BT.709-tagged clip beside an untagged one there needs a run on 8.0 (the decided plans for two untagged inputs stay).
• ProRes line, 8-bit source, overlay --image still composites through RGB.
• Text-default pictographs above U+1F300 (e.g. 🌡) are still counted as emoji; an emoji placeholder inside an italic cue ends with {\r}, so the text after it loses its italics.
• A truncated source reads as fine in 22 of 25 tools (tests/sweep.py on a one-third MKV): [from Velisa] Growing files and LTC #424's class.