PCA as post-processing

[AlekseySh]

1. Make an example on mock dataset and put it in the Docs here. It should look like this:

embeddings = inference(model, val, batch_size=4, num_workers=0)
rr = RetrievalResults.from_embeddings(embeddings, val, n_items=3)

embeddings_upd = PCA()(embeddings)
rr_upd = RetrievalResults.from_embeddings(embeddings_upd, val, n_items=3)

# after that we compare results visually and/or by metrics

2. Check if it boosts metrics on one of our benchmarks. Check how much the retrieval time is improved.

3. Write a shor report below the example on the same Doc's page.

PS. OML already has a PCA implementation, so, no need to bring an extra requirement.