docs: fix v0.2.0 release-audit documentation drift

[badbread]

Summary

Nine evidence-backed fixes from a v0.2.0 documentation/runbook audit. No code changes except forwarding SEGMENT_RECEIPT_TIMEOUT_SECS through compose (item 6) so an existing recorder log line's advice is actually true.

1. Migration range 0072-0078 -> 0072-0080 (docs/COMPOSE.md, docs-site/docs/getting-started/upgrade-and-rollback.md). Also reworded COMPOSE.md's "GO2RTC_AUTH is the only new key" to mention MAIN_REPAIR_TRANSCODE_ENABLED too.
2. Secrets docs overclaimed _FILE support for go2rtc credentials. GO2RTC_USER_FILE/GO2RTC_PASS_FILE don't work end to end (the embedded go2rtc reads the plain env vars directly, and compose :?-requires them). Scoped the _FILE-supported list to DATABASE_URL, JWT_SECRET, SEED_ADMIN_PASSWORD, HA_TOKEN in docs-site/docs/configuration/secrets.md and environment-reference.md.
3. docs/OPS-DEPLOY.md documented a nonexistent scripts/setup-env.sh --print flag. The parser only accepts --force/--prompt/-h. Replaced with the actual way to see the passphrase again (it's in .env as SEED_ADMIN_PASSWORD).
4. Stale DB_POOL_SIZE guidance. docs/OPS-DEPLOY.md said it needed hand-wiring into the environment blocks; the stock compose already forwards it and .env.example carries the commented key.
5. Backup-runbook contradiction. docs/OPS-DEPLOY.md and docs/OPS-BACKUP-RECOVERY.md only taught the manual scripts/backup-db.sh + cron path and never mentioned the api's built-in nightly pg_dump (on by default). Added a note to both framing the manual script as supplemental/ad-hoc.
6. Recorder log advised an env var compose never forwarded. services/recorder/src/recording.rs tells the operator to raise SEGMENT_RECEIPT_TIMEOUT_SECS, but it wasn't in the recorder's compose environment: block. Confirmed unset/unparseable safely falls back to the internal default (no panic, no zero-timeout) via clamp_segment_receipt_timeout_secs, so forwarding it is safe: added it to docker-compose.yml, .env.example, and the environment reference.
7. docs/ROADMAP.md "where we are today" was badly stale. The notification system described as "a single hardcoded webhook watchdog" is actually a full rules/channels/history engine (Discord, Slack, Pushover, Telegram, ntfy, webhook dispatchers; quiet hours; snooze/presence) shipped well before 0.2.0 and redesigned in the console by feat(notifications): redesign the console Notifications pane (channels, alert-text editor, quiet hours) #583. The distribution initiative claimed "no registry wired" and "CI does not build the APK", both shipped (public GHCR, signed Android releases via android-release.yml, version.properties at 0.2.0/5). Rewrote the "where we are today" passages and checked off the phase-checklist items that are actually done.
8. Plate naming was undocumented. Added a section to the LPR docs page covering naming a plate (distinct from watchlisting), that the name renders everywhere the plate appears, and copy-plate-number, shipped on all four clients (web, desktop, Android, iOS).
9. mainv documentation drift. docs/COMPONENT-MAP.md still documented the pre-fix(android): try the repaired main (mainv) before the doomed raw main #594 fallback order (main -> mainv -> ...); the code is mainv-first after fix(android): try the repaired main (mainv) before the doomed raw main #594. Also corrected the stale sdp_video_lacks_fmtp-on-producer detection detail to the actual served-SDP check (stream_served_video_lacks_fmtp, post-fix(api): flag the _mainv repair from the SERVED SDP, not the producer #592). Added a correction addendum to the relevant docs/DECISIONS.md entry (matching its existing addendum style) rather than editing the original decision text.

Test plan

• docker compose -f docker-compose.yml config -q validated clean on dev2 (real Docker, not a YAML parser) with the new SEGMENT_RECEIPT_TIMEOUT_SECS key rendering correctly
• No em-dashes/en-dashes introduced (checked the diff)
• Every fix cross-checked against the actual code/config it describes (see commit body for specifics)