codedb_remote MCP inputSchema underadvertises actions (4 of 11+)

[justrach]

Problem

The `codedb_remote` MCP tool implementation in `src/mcp.zig` supports 11+ actions and a `backend` parameter, but the `inputSchema` JSON only advertises 4:

\"action\":{\"type\":\"string\",\"enum\":[\"tree\",\"outline\",\"search\",\"symbol\"]}

The implementation actually accepts:

• codegraff: `tree, outline, search, meta`
• wiki: `tree, outline, search, symbol, policy, deps, score, cves, commits, branches, dep-history`

Plus a `backend` parameter (`codegraff` | `wiki`) and a `scope` parameter (`runtime` | `all`) for score/cves.

LLMs reading this schema can't discover ~7 actions that actually work. They learn about them only by trial-and-error or by reading docs separately.

Proposal

Update the inputSchema to:

• Enum the full action list
• Document the `backend` parameter explicitly
• Note action × backend matrix in the description, OR have a separate `actions` endpoint (see related issue)
• Add `scope` for score/cves
• Add the new params from the pagination issue (`limit`, `offset`, `prefix`, `since`)

Why

The schema is the only contract LLMs see. Hidden capabilities = unused capabilities.

[justrach]

Resolved by PR #340 — schema now enumerates all 13 actions, the backend × action matrix, and every named param.

[justrach]

Implemented by merged PR #340. The codedb_remote input schema now advertises the expanded action/parameter surface.