Fix MPS dtype mismatch in audio decoding and app relaunch on close

[Copilot]

Audio generation succeeded but decoding failed with invalid type: 'torch.mps.FloatTensor' on Apple Silicon. Additionally, closing the app caused an automatic relaunch.

Changes

Audio decoding dtype mismatch

• Frames generated on MPS with float16 were moved back from CPU without dtype conversion
• HeartCodec's detokenize() rejected float32 tensors when codec was loaded with float16
• Added explicit dtype conversion: frames.to(device=pipeline.codec_device, dtype=codec_dtype)

# Get codec dtype once for consistency
codec_dtype = getattr(pipeline, 'codec_dtype', torch.float32)

# Convert frames to codec device and dtype (critical for MPS float16)
frames_for_codec = frames.to(device=pipeline.codec_device, dtype=codec_dtype)
wav = pipeline.codec.detokenize(frames_for_codec)

App relaunch on close

• Exception handlers in launcher.py had infinite loops without interrupt handling
• Added KeyboardInterrupt handlers to allow clean shutdown in fallback mode

try:
    while True:
        time.sleep(1)
except KeyboardInterrupt:
    print("\nStopping server...")

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Decoding Audio Failure</issue_title>
<issue_description>After PR #14 generation succeeds at a reasonable speed, but once completed it fails with:

Deconding Audio:
Generation Failed: invalid type: 'torch.mps.FloatTensor'

Lets fix this last bug to make sure our generations are correctly decoded to audio.</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Decoding Audio Failure #15

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.