Switch to ROCm 6.1 + torch 2.5.1 to fix MIOpen workspace=0 slowness

ROCm 7.2 + PyTorch 2.11.0 has a bug where PyTorch passes workspace=0 to
MIOpen convolutions, forcing fallback to the slow GemmFwdRest solver.
This caused s3gen.inference to take 15-22s instead of <5s, making
synthesis 3-4x slower than real-time audio playback.

ROCm 6.1 allocates workspace correctly so MIOpen picks fast GEMM solvers
without needing torch.compile workarounds.

Changes:
- Base image: rocm/dev-ubuntu-22.04:7.2 → 6.1
- torch 2.11.0 → 2.5.1 (rocm6.1 wheel index)
- Add pytorch_triton_rocm==3.1.0
- transformers 5.2.0 → 4.46.3, safetensors 0.5.3 → 0.4.0
- s3tokenizer unpinned → 0.3.0
- resemble-perth==1.0.1 directly (v1.0.1 is pip-installable; drop stub)
- Drop Dockerfile perth_stub steps
- Drop torch.compile and timing patches from engine.py (not needed)
- Drop multi-pass warmup from main.py (torch JIT warmup not needed)
- Drop ROCm 7.2-specific env vars from docker-compose.yml

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

Dockerfile.rocm

@@ -1,4 +1,4 @@
-FROM rocm/dev-ubuntu-22.04:7.2
+FROM rocm/dev-ubuntu-22.04:6.1
 
 ENV DEBIAN_FRONTEND=noninteractive \
     PYTHONDONTWRITEBYTECODE=1 \
@@ -29,12 +29,6 @@ RUN pip3 install -r requirements-rocm.txt
 # Step 3: Install chatterbox with --no-deps so pip cannot replace ROCm torch.
 RUN pip3 install --no-deps chatterbox-tts
 
-# Stub out resemble-perth (audio watermarking, unnecessary for self-hosted use).
-# Placed after pip layers so changes to the stub don't bust the cache above.
-COPY perth_stub.py .
-RUN python3 -c "import site; print(site.getsitepackages()[0])" | \
-    xargs -I{} cp /app/perth_stub.py {}/perth.py
-
 # Application source
 COPY engine.py config.py wyoming_handler.py wyoming_voices.py main.py ./