feat(tts): Wyoming service discovery and multi-model seeding

[AsierZunzu]

Summary

• compose.yaml: Replaced the single tts service with tts-english and tts-spanish; removed PIPER_HOST/PIPER_PORT env vars from narratio and added PIPER_SERVICES: "tts-english:10200,tts-spanish:10200"; updated depends_on accordingly.
• src/utils/env.ts: Removed PIPER_HOST and PIPER_PORT accessors; added PIPER_SERVICES as a required string accessor (comma-separated host:port pairs).
• src/db/schema.ts: Added voice TEXT (nullable) and languages TEXT (nullable, stores JSON array) to the ttsServices table.
• src/db/index.ts: Updated TTS_SERVICES_SQL to include the new columns; added two ALTER TABLE migration statements for existing DBs; updated seedFromEnv to seed from PIPER_SERVICES instead of PIPER_HOST/PIPER_PORT.
• src/db/tts-services.ts: Added getTtsServiceByHostPort and upsertTtsServiceByHostPort functions; extended InsertTtsServiceParams with optional voice/languages fields.
• src/services/tts.ts: Added discoverService(host, port) — opens a TCP connection, sends {"type":"describe"}\n, reads until an info event, extracts tts[0].name (voice) and tts[0].languages, returns TtsServiceInfo | null. Returns null on any error (timeout, refused, malformed).
• src/worker/index.ts: Added seedTtsServices() — parses PIPER_SERVICES, calls discoverService for each endpoint, upserts into tts_services (updates voice/languages if the host:port row exists, inserts with a generated name otherwise), logs discovered voices, and warns about feeds referencing unknown TTS service IDs. Called once at startup before the first runOnce.
• tests/helpers/schema.ts: Added voice and languages columns to TEST_SCHEMA_SQL so in-memory test DBs match the updated schema.

New env var format

PIPER_SERVICES=host1:port1,host2:port2

Example: PIPER_SERVICES=tts-english:10200,tts-spanish:10200

Startup seeding flow

1. Worker starts → calls seedTtsServices()
2. For each host:port in PIPER_SERVICES: send describe → receive info → extract voice + languages
3. Upsert row in tts_services (insert on first run, update voice/languages on subsequent runs)
4. Log discovered voice at info level; log a warning for unreachable endpoints
5. Warn for any feed whose tts_service_id doesn't match a known service

Test plan

• npm run lint passes (tsc --noEmit, no type errors)
• npm test passes (103 tests across 11 test files)
• discoverService returns null gracefully on connection refused / timeout
• Existing DBs get voice/languages columns via ALTER TABLE migrations without data loss
• Fresh compose stack starts with two Piper services discovered and seeded

🤖 Generated with Claude Code