rotate: add an optional random rotation feature in INT8/INT4 quantization method

[zzlin237]

New Feature: Added an optional rotation parameter to INT8 quantization, now supporting HNSW, Flat, IVF, and Vamana indexes.
Background: The quantization error in INT8 is directly proportional to the data range (max - min). By applying a random rotation, we can distribute the variance evenly across all vector dimensions. This helps reduce the overall range in certain cases, thereby improving INT8 quantization accuracy.
Benefits: Recall is improved while maintaining the same QPS. Using the cohere-1m dataset as an example:

1. HNSW: QPS [14,431], Recall [0.9285] -> QPS [14,452], Recall [0.9397] (m=15, ef_construction=180, ef=180)
2. Flat: Recall [0.9695] -> Recall [0.9881]
3. Vamana: Recall [0.9576] -> Recall [0.9734]

Known Issues: We encountered issues during IVF testing and discovered pre-existing bugs in the IVF index:

1. The nprobe parameter at the Python layer is currently ineffective; the index defaults to a fixed scan_ratio=0.1.
2. When combining IVF with INT8, most points are assigned to a very small number of clusters (e.g., 70% of the data falls into a single cluster), which significantly degrades overall performance.

The test results for INT4 are as follows:

1. HNSW: Recall [0.2114] -> Recall [0.7117]
2. Flat: Recall [0.248] -> Recall [0.8146]
3. Vamana: Recall [0.2415] -> Recall [0.771]

[CLAassistant]

All committers have signed the CLA.

[feihongxu0824]

1. pr title需要能够说明这个pr的作用和目的；2. description里面需要注明一些背景、问题、收益等信息