[Feature] expose GPU/CPU choice and other parameters to pytorch cellfinder version

[alessandrofelder]

#440 adds some parameters to cellfinder image processing in Python. We've discussed on zulip how/which we'd like to expose these in the API (CLI and napari).

The plan is to

• do not expose use_scipy and set it always to True on CPU, and switch to pytorch when filtering on CPU slower in pytorch than scipy pytorch/pytorch#126115 is addressed (assuming this makes pytorch faster than scipy)?
• expose CPU/GPU choice (default should be GPU if there is one?)
• expose batch_size (and document its use well)

split_ parameters can be exposed later, if we know there is a good user base for this.