fix: relax dependency constraints to support Python 3.13

[ThanhNguyxn]

Summary

Fixes #22 - Current PyTorch dependencies don't work with CPython 3.13

Problem

The previous pyproject.toml had torch >= 2.2.0, < 2.5 which prevented installation on Python 3.13 because PyTorch only added CPython 3.13 wheel support starting from v2.5.0.

Solution

Relaxed dependency constraints to support Python 3.13 while maintaining backward compatibility with Python 3.10-3.12:

Dependency	Before	After
torch	>= 2.2.0, < 2.5	>= 2.2.0
numpy	>= 1.26, < 2.2	>= 1.26, < 3
safetensors	>= 0.4.0, < 0.5	>= 0.4.0
huggingface-hub	>= 0.24, < 0.25	>= 0.24
einops	== 0.7	>= 0.7
sentencepiece	== 0.2	>= 0.2
sounddevice	== 0.5	>= 0.5
sphn	>= 0.1.4, < 0.2	>= 0.1.4
aiohttp	>= 3.10.5, < 3.11	>= 3.10.5

Why These Changes Are Safe

1. Semantic Versioning: Most of these libraries follow semver, so minor/patch updates should be backward compatible
2. PyTorch Compatibility: The Moshi model architecture should work with newer PyTorch versions
3. Tested Range: Users with Python 3.10-3.12 will still get compatible versions from the existing ecosystem

Testing

This change enables installation on Python 3.13 by allowing pip to resolve torch 2.5+ which has the required wheels.

# Before: fails on Python 3.13
pip install git+https://github.com/NVIDIA/personaplex.git#subdirectory=moshi

# After: works on Python 3.13
pip install git+https://github.com/ThanhNguyxn/personaplex.git@fix/python313-compatibility#subdirectory=moshi

Backward Compatibility

✅ Fully backward compatible - no breaking changes for Python 3.10-3.12 users.