Populate flowsheet.dj_name on marker rows and surface it in v2

[jakebromberg]

Closes #952.

Problem

The v2 /flowsheet endpoint returned dj_name: "" on every show_start / show_end entry (and the same code path applied to dj_join / dj_leave). Verified live: 19/19 markers across a 48-hour window. iOS rendered the bare "Signed on/off" fallback; dj-site did the same.

Two-part root cause in apps/backend/services/flowsheet.service.ts:

1. startShow, endShow, createJoinNotification, createLeaveNotification inserted flowsheet rows without setting the dj_name column. The addEntry track path already did this (closed Populate flowsheet.dj_name on insert (step 5b.2) #476), but the marker paths were missed.
2. transformToV2 regex-parsed dj_name out of the human-readable message field instead of reading the column. The regex silently failed and returned ''.

Change

• Inserts: all four marker paths now persist dj_name using the same auth_user.djName || auth_user.name precedence as the existing track path and resolveDjNameForShow. Message strings unchanged for v1 consumers.
• Serializer: transformToV2 reads entry.dj_name directly for marker types; timestamp (show_start/show_end only) is derived from entry.add_time via toLocaleString('en-US', { timeZone: 'America/New_York' }) — same format the old regex was extracting from the message.
• Backfill: jobs/flowsheet-dj-name-backfill widens its filter from entry_type = 'track' to also cover the four marker types. Joins through shows.primary_dj_id — best-effort for historical guest dj_join/dj_leave rows (those will carry the primary DJ's name); new rows from the live insert path are exact.

TDD

Five steps, each Red → Green:

1. startShow / endShow persistence — tests/unit/services/flowsheet.startShow.test.ts, flowsheet.endShow.test.ts (new).
2. dj_join / dj_leave persistence — tests/unit/services/flowsheet.joinLeaveNotifications.test.ts (new).
3. transformToV2 reads from column — rewrote the marker describe blocks in flowsheet.service.test.ts to assert column read and ignore message content.
4. Backfill widening — updated tests/unit/jobs/flowsheet-dj-name-backfill/job.test.ts to assert the new entry_type IN (...) filter.
5. Doc sync — CLAUDE.md entry, file-header comments in flowsheet.service.ts (L755-761) and job.ts.

Final: 67/67 tests green on the modified suites. (Pre-existing failures in library.service.test.ts, flowsheet.controller.test.ts, internal.route.test.ts are unrelated — LRUCache import + network socket binding; they fail on main too.)

Verification plan

• npm run test:unit (all modified suites green; pre-existing failures unchanged).
• npm run test:integration (no regression).
• Deploy to staging.
• Run backfill: docker run --rm --env-file .env <flowsheet-dj-name-backfill image>. Confirm verifyComplete reports 0 missing.
• curl -s "https://staging-api.wxyc.org/flowsheet?shows_limit=10" | jq '.entries[] | select(.entry_type | startswith("show_") or startswith("dj_")) | {entry_type, dj_name, timestamp}' — every marker has a populated dj_name.
• iOS smoke: scroll the playlist past a sign-on/off boundary, confirm "<DJ> signed on/off" renders.
• Production deploy + backfill.

Out of scope

• Perfect attribution for historical guest dj_join/dj_leave rows (acknowledged limitation; follow-up if needed).
• Flowsheet ETL: shows.legacy_dj_name occasionally not populated even when tubafrenzy has DJ_NAME #605 (shows.legacy_dj_name ETL gap) and flowsheet-etl: post-import dj_name UPDATE crashes on large legacy_entry_id batches (400+ param ANY clause) #939 (large-batch ANY((...)) crash) — separate bugs. This fix is robust to both: missing legacy data inserts dj_name = null and the next backfill picks it up.