feat: Zvec / zvec-grep (zg) backend for semantic codebase_search (#4066)

## Context & Proposal

Alibaba open-sourced [zvec](https://github.com/alibaba/zvec) (an in-process vector DB built on Proxima, Apache-2.0) and [zvec-grep](https://github.com/zvec-ai/zvec-grep) (`zg`: unified ripgrep + BM25 + vector search CLI with local-first posture & MCP surface).

This fits issue #4066 (`feat: semantic codebase_search tool`), currently config/icon-only:
- **Local-first hybrid search**: `zg` unifies exact/regex, BM25, and local vector search into ranked, source-linked results.
- **Provider neutrality & DashScope integration**: Combines with local embedding models or Qwen embeddings via DashScope API key.
- **Token efficiency**: Reduces tool calls by delivering line-level cited evidence directly for agent consumption.

## Spike Plan
1. Evaluate `zg` CLI & MCP adapter in local environment.
2. Prototype `codebase_search` tool wrapper invoking `zg query` / MCP server.
3. Compare search accuracy, latency, and context budget efficiency vs BM25 tool_search.

Ref: #4066

No implementation files or tests are named. Start by evaluating the zvec-grep CLI and MCP adapter locally, then prototype a codebase_search wrapper invoking `zg query` or the MCP server. Done means comparing accuracy, latency, and context-budget efficiency with the existing BM25 tool_search approach.

Written by the indexing model from the issue text.