Fix IVTC failure on >8-bit sources (e.g. 10-bit ProRes 422)

[StuartCameronCode]

Problem

A user reported that the preview (and, in fact, the full encode too) fails on a 10-bit ProRes 422 source:

Python exception: VFM: input clip must be constant format YUV420P8, YUV422P8, YUV440P8, YUV444P8, or GRAY8
  ...
clip = core.vivtc.VFM(clip, order=0, ...)

The source decodes to yuv422p10le (10-bit). This isn't a global 8-bit limitation — it's specific to the IVTC (Inverse Telecine) pass: vivtc.VFM only accepts 8-bit YUV/GRAY. descratch already had an 8-bit guard, but the IVTC pass had none — in both preview_template.vpy and pipeline_template.vpy, so the actual export would have failed identically, not just the preview.

Fix

The IVTC block now runs VFM's field-matching on an 8-bit metrics copy (only converted when the source isn't already 8-bit) while emitting the full-depth pixels via VFM's clip2 parameter — so 10-bit precision is preserved rather than truncated to 8-bit.

Confirmed against the bundled libvivtc that VDecimate already accepts 8–16 bit natively (only VFM is 8-bit-limited), so VDecimate runs directly on the full-depth VFM output and the soft-telecine path needed no change.

Test fixture

Adds Tests/TestResources/prores422_10bit_telecine.mov — a small (2.9 MB, 30-frame) Apple ProRes 422 Standard telecine clip (yuv422p10le) matching the reported source, plus generate_prores422_10bit_test.sh to reproduce it. It can be dropped into the app to exercise the high-bit-depth IVTC path via the UI.

Tests

• Script-gen (push gate):
  • worker/tests/filter_integration_test.rs → test_57_ivtc_high_bit_depth_guard
  • app/test/integration_filter_parameters_test.dart → ivtc: VFM/VDecimate 8-bit guard + clip2 (adds a deinterlace param to the test's buildJob helper)
  • Both assert the guard + clip2 wiring appear in the generated script.
• End-to-end regression guard (nightly / heavy):
  • app/test/integration_high_bit_depth_test.dart → probes the ProRes fixture, confirms it is >8-bit, then runs the worker end-to-end with IVTC and asserts a valid encode. Verified it fails on the pre-fix template (encode aborts at the VFM 8-bit error) and passes with the fix — the real regression guard the script-only tests cannot provide.

Manual verification

Decoded the 10-bit yuv422p10le source through the real ffmpeg | vspipe | ffmpeg pipe: previously errored, now processes cleanly (90→72 frames) with the output remaining yuv422p10le.

🤖 Generated with Claude Code