feat(search): HNSW New Arguments

[PavelPashov]

Description

RediSearch 8.12 adds SQ8 scalar quantization to HNSW vector fields. Add two options to SchemaHNSWVectorField:

• COMPRESSION: 'SQ8' (TYPE must be FLOAT32 or FLOAT16)
• TRAINING_THRESHOLD: number (0-102400). Applies only with COMPRESSION. 0 disables mean normalization, so an explicit 0 is sent to the server.

---

Checklist

• Does npm test pass with this change (including linting)?
• Is the new or changed code fully tested?
• Is a documentation update included (if this change modifies existing APIs, or introduces new ones)?

---

Note

Low Risk
Additive API and command serialization for optional HNSW index parameters, with tests only; no changes to auth, data paths, or existing index behavior.

Overview
Adds RediSearch 8.12 HNSW vector options to client.ft.create: COMPRESSION: 'SQ8' (scalar quantization for FLOAT32 / FLOAT16) and TRAINING_THRESHOLD (0–102400; only meaningful with compression, including explicit 0).

CREATE now maps these fields on HNSW schemas to the matching FT.CREATE tokens (COMPRESSION, TRAINING_THRESHOLD), with TypeScript types and JSDoc on SchemaHNSWVectorField.

Tests cover argument transformation and, on Redis ≥ 8.12, successful index creation/FT.INFO round-trip plus server rejection when TRAINING_THRESHOLD is set without COMPRESSION.

^{Reviewed by Cursor Bugbot for commit d5c5b5d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[nkaradzhov]

LGTM