Refactor of PyTorch data loading classes and methods

[Natooz]

Following the discussions in #147, this PR implements a new PyTorch Dataset class, with the particularity of splitting the MIDI itself instead of splitting the tokens as done previously.

---

📚 Documentation preview 📚: https://miditok--148.org.readthedocs.build/en/148/

[codecov]

Codecov Report

Attention: Patch coverage is 92.96375% with 33 lines in your changes are missing coverage. Please review.

Project coverage is 91.24%. Comparing base (408eec0) to head (48055b4).

Files	Patch %	Lines
miditok/pytorch_data/split_midi_utils.py	89.76%	13 Missing ⚠️
miditok/utils/utils.py	92.81%	11 Missing ⚠️
miditok/pytorch_data/datasets.py	92.63%	7 Missing ⚠️
tests/test_pytorch_data_loading.py	96.36%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #148      +/-   ##
==========================================
+ Coverage   91.08%   91.24%   +0.15%     
==========================================
  Files          33       34       +1     
  Lines        4994     5299     +305     
==========================================
+ Hits         4549     4835     +286     
- Misses        445      464      +19     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.