Add startup warmup and make fp16 autocast fault-tolerant

Warmup: run a synthesis before accepting Wyoming connections so MIOpen
benchmarks and caches all conv layer shapes. Without this, the first HA
request triggers hundreds of benchmark runs and times out.

fp16: wrap in try/except so a failed autocast retries in fp32 rather
than dropping the request silently.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

main.py

@@ -18,6 +18,17 @@ logging.basicConfig(
 logger = logging.getLogger(__name__)
 
 
+def _warmup(voices: dict) -> None:
+    from wyoming_voices import resolve_voice
+    audio_prompt = resolve_voice(None, voices) if voices else None
+    logger.info("Running warmup synthesis to populate MIOpen kernel cache...")
+    try:
+        engine.synthesize(text="Warmup.", audio_prompt_path=audio_prompt)
+        logger.info("Warmup complete — MIOpen cache populated")
+    except Exception:
+        logger.warning("Warmup synthesis failed (non-fatal)", exc_info=True)
+
+
 async def main() -> None:
     load_config()
 
@@ -29,6 +40,12 @@ async def main() -> None:
     voices = load_voices()
     wyoming_info = create_wyoming_info(engine.get_sample_rate(), voices)
 
+    # Run a warmup synthesis before accepting connections so MIOpen benchmarks
+    # and caches the best convolution algorithms for all layer shapes. Without
+    # this, the first real HA request triggers benchmarking (hundreds of runs)
+    # and times out before any audio is returned.
+    _warmup(voices)
+
     host = get_wyoming_host()
     port = get_wyoming_port()
     uri = f"tcp://{host}:{port}"