feat(mcp): optional limit on callers, callees, and impact

[solomonneas]

What

Adds an optional limit argument to the callers, callees, and impact MCP tools. When set, the edge list is cut to the first N real edges and one marker row is appended reporting shown vs total (results limited to 5 of 79 edges). When omitted, output is byte-for-byte identical to before.

Why

A tool-surface benchmark (receipts in benchmarks/tool-surface/) measured the typed tools against a single bounded-search surface on the same 12-question set. Typed tools won on answers (11/12 vs 4/12) but had no per-call size lever: impact on a hot symbol emitted ~9,150 tokens and callers ~7,460 in one call, against the internal 500-per-direction safety cap only. An agent on a token budget needs to bound a blast-radius query the same way it bounds a search.

Measured on this repo: callers sync_repo unbounded = 79 edges / 33,479 chars; limit: 5 = 5 edges + marker / 2,267 chars.

How

• limit_edges(edges, Option<usize>) in src/query/graph.rs: pure function, None passthrough, counts only real edges, drops any pre-existing cap marker on truncation, appends one TRUNCATED_EDGE_KIND sentinel.
• Wired in src/mcp.rs dispatch for the three tools; limit added to the shared symbol-tool schema (optional) and validated like depth (absent ok, non-integer rejected).
• Unit test limit_edges_truncates_and_reports_shown_vs_total.

Verification

cargo fmt --check, cargo clippy --all-targets --all-features -- -D warnings, cargo test --all-features, cargo build --release all pass (Brigade receipts 20260717-1926*). Live MCP session confirmed omitted-limit output unchanged and limit: 5 behavior above.