TT-7621 test: PBT CT harness and defect repro coverage

[nabalone]

This PR is now tests-only for the TT-7621 PBT work.

What is in this PR

• Cypress CT harness that mounts the real PBT step (real wavesurfer, MediaRecord flow, Orbit memory source, UnsavedProvider, and upload action chain)
• PBT component specs for behavior, editing, selection, and known-defect repros
• Known-defect tagging so normal CT runs stay meaningful while defect repros remain runnable
• CT support updates (--mute-audio, node tasks for CT, pre-bundled harness deps, CT-specific Vite cache)

What is not in this PR

The segment-selection state fix that was previously bundled here has been split to:

• TT-7621 fix: follow a segment change the waveform reports #536 (TT-7621_pbt-follow-segment-change)

Notes

• Base: develop
• Stack: other TT-7621 fix PRs are based on this branch

[nabalone]

0-based vs 1-based currentSegmentIndex — background, and why this change token exists.

The field has two conventions in active use:

	Convention	Where
Writes	1-based (sortedIndex + 1, 0 = whole)	usePlayerLogic.ts:113 (the waveform), Mark Verses (table row — row 0 is the header)
Writes	0-based clause index	PassageDetailGuidedPhraseRecord (17 call sites)
Reads	1-based	PassageDetailItem.tsx:246 (filename postfix sN), DiscussionList.tsx:107, Mark Verses row lookup
Reads	0-based	useWavesurferRegions.tsx:955 — regarray[defaultRegionIndex]?.start

Nothing forces agreement: setCurrentSegment(segment, index: number) accepts any number, so each author picked what was natural locally. That is how a genuine move can arrive carrying the number the previous writer used, which is the bug this commit fixes.

Noel's call: unify the numbering after the next release — it touches every caller across Mark Verses, the players and the guided-record step, which is more than we want to move right now.

When that happens, this token becomes unnecessary: with one convention the index is a reliable change signal for this case, and the two dependency lists in PassageDetailGuidedPhraseRecord can go back to watching currentSegmentIndex alone. One caveat to check then — same sorted position with different bounds (after Combine, position 1 spans 0–6 where it spanned 0–3) still repeats the number. No reproduction for it today; the token covers it for free.

That last table row is also a latent off-by-one worth its own ticket: loadRegions ends with onRegionGoTo(regarray[defaultRegionIndex]?.start ?? 0) and applyRegionAtPosition then selects the region at that position — so a region load reading a 1-based value selects the segment after the current one. loadRegionsJson is called by Split, Combine, segment undo and Reset. PBT happens to mask it (an effect there keeps re-asserting its own 0-based index), and I could not produce a user-visible failure from that screen — but Mark Verses and the generic segment player have no such effect.

[nabalone]

This dependency list is the concrete thing that becomes removable once the 0-based / 1-based numbering is unified — see the note on PassageDetailContext.tsx.

After unification (Noel: after the next release), both navigation effects can drop currentSegmentSeq and watch currentSegmentIndex alone, and the 17 setCurrentSegment(region, idx) calls in this file send whatever the agreed convention is.

Worth knowing while reading this effect: the step never uses the number — it derives its own idx from the region via findClauseIndex. The index was only ever here as a change signal, which is why it was so easy for the two numbering schemes to collide silently.