Allow for specifying a tensor device in AnnLoader

[austinv11]

Currently AnnLoader only supports transferring tensors to cuda devices. However, PyTorch as of late has much improved its support of other devices such as mps (Mac GPUs) and XLA (Google TPUs). So I added a device argument that is more semantically similar to other PyTorch APIs to defer to the user to decide which accelerator device (if any) to load the data onto.

• Closes N/A
• Tests added: There are currently no tests for AnnLoader. So I am not sure how to add tests for this.
• Release note added (or unnecessary)

[codecov]

Codecov Report

Merging #1240 (91f66fb) into main (3e340e1) will decrease coverage by 2.27%.
The diff coverage is 22.22%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1240      +/-   ##
==========================================
- Coverage   85.27%   83.00%   -2.27%     
==========================================
  Files          34       34              
  Lines        5404     5408       +4     
==========================================
- Hits         4608     4489     -119     
- Misses        796      919     +123     

Flag	Coverage Δ
gpu-tests	?

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
anndata/experimental/pytorch/_annloader.py	22.11% <22.22%> (+0.11%)	⬆️

... and 8 files with indirect coverage changes