fix(export): normalize meeting transcript timestamps for markdown exports (Fixes #1472)

[stantheman0128]

Summary

• Normalize epoch-millisecond meeting segment timestamps to recording-relative offsets before TXT/SRT/JSON/Markdown export.
• Parse SQLite UTC note timestamps explicitly so export headers render in local time instead of appearing one hour behind BST.

Root cause

Diarization finalization already subtracts recording start before merge, but export-transcript read raw
ote.transcript JSON and passed epoch-ms values into ormatTimestamp(), which expects seconds from zero. That produced giant hour fields like 496069042:26:40.

Verification

\
cd openwhispr-wt/1472-markdown-export-timestamps
node --test test/helpers/transcriptFormatter.test.js

13 pass (includes epoch-ms normalization + UTC created_at parsing)

\\

AI assistance

AI-assisted (Cursor). Stan Shih reviewed the diff.

Fixes #1472

Made with Cursor

[gabrielste1n]

Merged as #1560 — your fix, unchanged, rebased onto current main. Thank you @stantheman0128, this was a clean diagnosis of two genuinely separate bugs.

Why it moved branches rather than being rebased in place: my token can't push to a fork branch when the rebase brings .github/workflows/ changes along with it, which any rebase onto current main now does. Nothing to do with your work.

The only conflict was in test/helpers/transcriptFormatter.test.js and it was purely additive — main picked up i18n speaker-label tests from #1426 while yours added the timestamp tests. Both sets are in; nothing was dropped.

Worth recording that the blast radius was wider than the title suggests: formatTxt, formatSrt, formatJson and formatMd all share mergeSegments/formatTimestamp, and markdownMirror.js:110 writes through formatMd — so every export format and the on-disk mirror were affected, not just manual markdown export. Your fix covers all of them.

[stantheman0128]

Thanks @gabrielste1n — glad the diagnosis held. Appreciate you landing it as #1560 when the fork branch was awkward to push.