Improve Hnsw cost estimate

[Ngalstyan4]

Properly implement postgres index cost estimation for postgres query planner

Resources

• Index cost estimation` https://www.postgresql.org/docs/current/index-cost-estimation.html

[yoloVoe]

From Narek:

That issues requires quite an in depth understanding of the hnsw algorithm. You need to understand how the hyper-parameters of the algorithm (used in index creation via CREATE INDEX) affect the number of index pages read. So, make sure you read the hnsw paper and play with LanternDB before starting with this issue.

[yoloVoe]

Commits that make progress on this task.

• First commit https://github.com/yoloVoe/lanterndb/commit/bfebc9f513f71304652ae43f4cfdf5d3bd6b1076 essentially just adds some logs to current implementation so we can track of some relevant internal state in existing tests.
  If we don't want this in non-test builds, maybe we can add a compiler flag/macro that only enables it during test runs.
  This commit can be submitted as PR already.
• Second commit https://github.com/yoloVoe/lanterndb/commit/0dfc36a6b5a6de3cddf67ead6365a5a5d8a0982f improves the hnsw cost estimate. Some items that need to be done before this can be submitted as PR: 1) Update tests, 2) Sanity-check values around edge-cases (LIMIT > number of vectors in index, etc)

There are further improvements we can do on top of the implementation above, but to gain confidence in improvements, we'd also need to add more test cases. Ideally, we'd be able to unit test the cost estimate function too.