Create Cache class for exact, fuzzy, and semantic deduplication

[sarahyurick]

TODO:

• Exact deduplication files
• Semantic deduplication files
• Fuzzy deduplication files
• Tutorials folder

[sarahyurick]

This PR allows the user to use Cache(cache_dir=...) instead of having to specify the cache_dir at every step of the deduplication pipelines.

However, the user does not have to use Cache if they want to keep using the modules the same way as before. Both ways work, with preference if the user specifies the cache_dir field in a deduplication module.

The only hard changes are with the semantic deduplication fields, see below.

[sarahyurick]

We deprecate clustering_output_dir in favor of cache_dir and clustering_save_loc for ClusteringModel, which matches the logic in the SemDedup class.

[sarahyurick]

We deprecate embedding_output_dir in favor of cache_dir and embeddings_save_loc for EmbeddingCreator, which matches the logic in the SemDedup class.

[sarahyurick]

We deprecate emb_by_clust_dir and sorted_clusters_dir in favor of cache_dir and embeddings_save_loc for SemanticClusterLevelDedup, which matches the logic in the SemDedup class.

[Maghoumi]

Thanks so much for working on this change. What I like about this now is that it gives users the option to either use the same cache directory for anything that requires caching, or provide a specific directory if they don't want to re-use the same cache.

The cache class implementation is functional but not thread-safe. I don't think that's a blocking problem for this PR.

I didn't run the samples/tutorials, but I assume the change has been thoroughly verified?

[sarahyurick]

Thanks so much for working on this change. What I like about this now is that it gives users the option to either use the same cache directory for anything that requires caching, or provide a specific directory if they don't want to re-use the same cache.

The cache class implementation is functional but not thread-safe. I don't think that's a blocking problem for this PR.

I didn't run the samples/tutorials, but I assume the change has been thoroughly verified?

Thanks! Yes, I tried to make as few breaking changes as possible. The examples and tutorials should all reflect those changes.

[sarahyurick]

IIRC, I removed seed and largest_cluster_size_to_process from the semantic dedupe parameters because I couldn't find them being used anywhere.