feat: setting up sae_lens package and auto-deploy with semantic-release

[chanind]

This PR renames the package to sae-lens to ready it for publication on PyPI. All imports are now from sae_lens. The sae_training and sae_analysis packages are now accessed at sae_lens.training and sae_lens.analysis, respectively. The most commonly imported items from the training package are also exported at the top-level, so it's possible to run from sae_lens import SparseAutoencoder, for example.

In addition, this PR sets up the library to auto-deploy on commit to main based on semantic-release. This means that the commit message to main will determine how the version is bumped in a deploy, or whether a deploy happens at all. For example, the following commit messages will bump the version number as follows:

• chore: updating docs no version bump
• fix: the main class no longer errors on osx patch version bump (1.2.3 -> 1.2.4)
• feat: added a new function minor version bump (1.2.3 -> 1.3.0)
• feat: redid everything BREAKING CHANGE: everything is different major version bump (1.2.3 -> 2.0.0)

I tried to update all the docs and jupyter notebooks to reflect the new import pattern, but it's possible I missed something somewhere.

This PR adds a custom unpickler method so that currently existing pickled SAEs on huggingface can still be loaded with torch.load() in the session loader.

closes #10

[codecov]

Codecov Report

Attention: Patch coverage is 67.21311% with 20 lines in your changes are missing coverage. Please review.

Project coverage is 55.18%. Comparing base (b8fba4f) to head (3ce7f99).

Files	Patch %	Lines
sae_lens/training/utils.py	38.09%	13 Missing ⚠️
sae_lens/training/toy_model_runner.py	0.00%	4 Missing ⚠️
sae_lens/training/sae_group.py	84.61%	1 Missing and 1 partial ⚠️
sae_lens/training/train_sae_on_toy_model.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #66      +/-   ##
==========================================
+ Coverage   52.77%   55.18%   +2.41%     
==========================================
  Files          14       15       +1     
  Lines        1260     1272      +12     
  Branches      198      198              
==========================================
+ Hits          665      702      +37     
+ Misses        553      530      -23     
+ Partials       42       40       -2     

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

[jbloomAus]

Amazing, thanks so much!