February bugfixes/improvements!

[casblaauw]

• Adds an info message when lazily loading a module, to inform the user when this happens (fixes Inconsistent lazy loading with crested.pl #165)
  • I've been confused by very slow plotting functions that randomly decided they want to lazily load .tl, loading Keras, which can take like 5 min with my setup. This informs the user what's going on.
• Adds a copy argument to all pp functions, which returns a copy of the anndata instead of modifying it in-place. Analogous to scanpy's copy.
• Allow single string chromosomes in pp.train_val_test_split, in my eternal quest against requiring single-length lists for single entries.
• Delete .tar files of datasets/models after untarring them.
• Test that tl.predict correctly takes strand into account.
  • This one is just because an unexpected change to this made me think my models were broken for like a month.
• Fix docs building since it's incompatible with the pandas 3.0.0 docs website for now (fixes Creating the docs breaks because of pandas' docs being updated #181)

[casblaauw]

Ugh, this has the same readthedocs problems as I detailed in #166. I'm pinning sphinx to below 9 for here now and will undo it in #166 (since by the time we merge that, I'm expecting myst-nb to have released their fix).

[casblaauw]

I'm so tired of CIs that randomly decide to break without having to do anything with your code. What do you mean you're trying to refer to pandas objects (where???) and that breaks now without ever breaking before?

[casblaauw]

Never mind, see #181 for the reason the docs building was broken. Fixed for now by pointing pandas' intersphinx link to an older version.