Add coderag query verb: NL search + structured filters over the index

[ryan75195]

Implement a `coderag query` CLI verb that consumes a `.coderag/index.db` produced by `coderag index` and returns matching chunks ranked by embedding similarity to a natural-language query.

Scope

• `coderag query [--db ] [--top-k N] [--format text|json] [--kind ] [--project ] [--namespace ] [--is-async]`
• Embed query with the same model the index used (`text-embedding-3-large`, 3072 dim)
• Apply structured filters (kind/project/namespace/is-async) BEFORE KNN where possible to narrow the search
• Run KNN against `chunk_embeddings` virtual table
• Default output: text with `relative_file_path:line_start fully_qualified_symbol_name score` and the chunk source body
• `--format json` emits machine-parseable output

Out of scope (this issue)

• Re-ranking with BM25 or other hybrid approaches
• MCP server interface (separate issue)
• Filters across child tables (`chunk_attributes`, `chunk_method_parameters`)