Benchmarking and documentation improvements towards ann-benchmarks integration

[alexklibisz]

Running ann-benchmarks is still a bit flaky, but this PR includes a bunch of useful improvements to the general benchmarking setup.

External-facing changes:

• Renamed parameter r in L2Lsh mapping to w, which is more appropriate and common for "width".
• Updates and fixes in the Python client based on usage for ann-benchmarks. Mainly adding/fixing data classes in elastiknn.api.

Internal changes:

• Removed continuous benchmark github workflow. Had to constantly tweak it to make it fit under 7gb memory (especially after using tmpfs). Also merging segments takes abnormally long on the GH workers compared to locally.
• Simplified benchmark app CLI params.
• Benchmarking argo workflow and local docker-compose use in-memory (i.e. tmpfs) storage for the Elasticsearch. This mainly speeds up indexing: exact glove25 drops from 186s to 114s, exact sift drops from 249s to 123s. No obvious improvement in query speed, likely because the OS is caching access to the Lucene files anyways.
• Python client uses stored field for document ID.
• Re-wrote results report script to generate results in markdown which can be easily copied into benchmarks.md doc page.
• Documentation improvements.