Port Omni Flash text-to-video to migrated Flow API

[Bl0ck154]

Summary

Ports Gemini Omni 1.1 Flash text-to-video onto the migrated flow.google.com batchexecute transport.

The migrated editor submits Omni T2V through RPC YhhmEf; duration is encoded in the model key (abra_t2v_4s, abra_t2v_6s, abra_t2v_8s, abra_t2v_10s). Completed media is resolved through the existing migrated as29s lookup.

This PR intentionally does not claim migrated support for Omni first-frame, start+end-frame, or reference-to-video. Those modes remain fail-fast until their new payloads are captured.

What changes

• add the YhhmEf request builder/parser to flow_batch.py
• add generate_omni_flash_text_video() for 4/6/8/10 second T2V
• expose POST /api/flow/generate-video-omni-text
• make /api/flow/check-omni-status resolve migrated Omni media through as29s
• keep frame/reference Omni modes explicitly unsupported on the batch path
• update docs so migrated and legacy Omni capabilities are not conflated

Live verification

Verified against the current migrated Flow service with a real paid generation:

• RPC: YhhmEf
• model: abra_t2v_4s
• prompt: simple red paper boat motion test
• Flow accepted the request and returned a real media/workflow record
• media resolved through as29s
• downloaded MP4 verified with ffprobe: 1280×720, H.264, 24 fps, exactly 4.000 s

A rejected 4-second Veo lower-priority model probe returned PUBLIC_ERROR_MODEL_ACCESS_DENIED before generation; no blind model retries are part of this implementation.

Validation

• 82 passed across test_flow_batch.py and test_omni_flash.py
• MV3 extension cold-start regression test passes
• git diff --check clean

The branch is based directly on current upstream main (ce37f4b) and contains only the generic migrated Omni T2V work.