import libemg now unconditionally loads torch + tslearn (PyQt6 conflicts)

[Michiboi29]

Since #130 (2c21236), libemg/init.py ends with from libemg import _discrete_models, whose submodules import torch and tslearn at module level.
As a result, any import libemg, even just from libemg.data_handler import OnlineDataHandler, now loads torch.

Reproduction
pip install libemg
python -c "from libemg.data_handler import OnlineDataHandler; import sys; print('torch' in sys.modules)"

True (was False pre-#130)

Results

• torch and tslearn are now hard install deps for users who never touch discrete models.
• import libemg start-up time grew significantly (torch init).
• On Windows, any consumer that imports a Qt binding before libemg now hits torch's well-known DLL conflict (Qt's Qt*\bin shadowing the runtimes
  c10.dll needs → WinError 1114). Pre-Latest #130 consumers were insulated because libemg didn't load torch.

Possible fix :

• Drop the from libemg import _discrete_models line — let users opt in with from libemg._discrete_models import ….

• Lazy-import torch/tslearn inside the discrete-model classes rather than at module top.

• Move torch / tslearn to an optional extra (libemg[discrete]).

  Happy to PR whichever direction maintainers prefer.