Improve DiskANN Hybrid Search via Alpha-Strategy

[PwzXxm]

issue: #142
/kind enhancement

This PR contains:

1. Implement alpha-strategy in DiskANN cached_beam_search, which makes at most 6x QPS improvements, 87% recall at filter rate 90%.
2. Only compute PQ distances for unvisited/unfiltered neighbors instead of all neighbors. This yields around 16% and 34% improvements in QPS on Cohere 1M and OpenAI 1M on the test machine. This result might not be convincing as Milvus might not be stable at the time of the test. Might need to investigation further.
3. Remove the dynamic threshold and fix it to 0.93 as HNSW.

The Milvus E2E performance is shown below.

Note that in the above figure, the recall at the filter rate of 95% does not match with the code, only for experimental purposes.

[sre-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: PwzXxm

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [PwzXxm]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[alwayslove2013]

/lgtm