Fix warmup text length and ve attribute for torch.compile

- Warmup now uses a ~170-char representative sentence so torch.compile
  JIT-compiles for typical token sequence lengths. Previously "Warmup."
  compiled for very short shapes, causing a full re-compile (17s) on the
  first real HA request and pushing total synthesis past 30s.
- Compile model.ve (voice encoder) in addition to s3gen — both are
  convolutional and hit the MIOpen workspace=0 bug.
- Fix _patch_timing: attribute is model.ve not model.voice_encoder,
  so the timing wrap was silently skipping the speaker embedding.

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

main.py

@@ -23,7 +23,13 @@ def _warmup(voices: dict) -> None:
     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)
+        engine.synthesize(
+            text=(
+                "This is a warmup synthesis request used to pre-compile neural network kernels "
+                "for typical text lengths, so that the first real request runs at full speed."
+            ),
+            audio_prompt_path=audio_prompt,
+        )
         logger.info("Warmup complete — MIOpen cache populated")
     except Exception:
         logger.warning("Warmup synthesis failed (non-fatal)", exc_info=True)