feat: local TTS via piper subprocess with Google fallback

- New src/tts.rs: centralize TTS in synthesize(config, http_client, text)
  Piper subprocess primary, Google Translate fallback on error/missing
- TTS_MODEL_PATH/TTS_VOICE env config (defaults to pl_PL-bass-high)
- Fix Unknown interaction errors: spawn play_tts_in_channel in background
  so voice-state TTS no longer blocks the shard event loop (3s Discord limit)
- Best-effort confirmation replies in /tts to prevent error cascade
- Custom on_error handler: graceful interaction-expired error handling
- Comprehensive tracing logging (engine, model, duration, stderr)

1 files changed, 2 insertions, 1 deletions

diff --git a/AGENTS.md b/AGENTS.md
index f5a78ee..a074a43 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -15,7 +15,7 @@ src/
 │   ├── music.rs         — /play (🟡 stub), /stop, /join, /loop_toggle
 │   ├── qr.rs            — /qr encode + decode
 │   ├── trigger.rs       — /trigger group (6 subcommands)
-│   ├── tts.rs           — /tts (Google TTS → file/vc), /log_volume
+│   ├── tts.rs           — /tts (local Piper TTS → file/vc), /log_volume
 │   ├── utilities.rs     — compress🟡, convert🟡, upload, search, download, say
 │   └── web.rs           — /web (DuckDuckGo scrape)
 ├── copyparty/mod.rs     — CopypartyClient (search, upload)
@@ -23,6 +23,7 @@ src/
 ├── events/mod.rs        — handle_ready, handle_voice_state_update, handle_message
 ├── openrouter/mod.rs    — OpenRouterClient (chat, chat_with_history)
 ├── tasks/mod.rs         — cron_scheduler_loop (30s polling, stub exec bodies)
+├── tts.rs               — synthesize(): Piper subprocess → Google TTS fallback
 └── voice/mod.rs         — VoiceManager (connect, disconnect, get_call), mp3_to_input
 ```