Fix/audio srt timeline 216

[Minh3132]

Summary

• TODO

User impact

Who benefits from this change, and what FunClip workflow improves?

Validation

• I ran the relevant tests or commands:
• I checked the affected README/docs links, if changed.
• I verified the affected UI, API, ASR, LLM, or clipping path, if changed.

Screenshots or clips

Add before/after screenshots, subtitles, timestamps, or short clips for UI or clipping-result changes.

Notes for reviewers

Mention any follow-up work, known limitations, or files that deserve extra attention.

[LauraGPT]

Thanks for working on #216. I checked head 91977f6: both submitted tests pass, but an empty first slice still shifts the following subtitle earlier while leaving the audio correct. This overlaps with the existing #217; its current b00e900b passes the same case. Please coordinate around that implementation instead of reintroducing the earlier accumulation bug, and fill in the PR summary/validation rather than leaving the template.

Bounded validation used the exact candidate VideoClipper module and hash-verified base helpers with real NumPy/subtitle code, synthetic audio and no ASR/model inference. This is not full-suite, browser/video or accuracy validation.

[LauraGPT]

[P2] Accumulate actual appended samples, not the signed interval length. With 8 s of 16 kHz audio, timestamp_list=[[16000,32000],[64000,112000]], start_ost=1500, end_ost=0, and sentence timestamps [[1000,2000],[6000,7000]] ms, the first clamped slice is 2.5–2.0 s and therefore empty. This line adds -0.5 s anyway. The second slice contributes 24000 correct samples, but its world subtitle becomes 0.000–1.000 s instead of 0.500–1.500 s. I reproduced the correct timing on base and current #217, and the early timing on this head. Use an integer count of samples actually appended, deriving seconds only when calling the subtitle helper, in both accumulation sites; add empty-first/middle and offset/clamping regression cases.