Agent-friendly interface: CLI + parser/analysis skills

[junjzhang]

Expose memviz/neo's parser and analysis primitives outside the browser UI so code agents (Claude Code, etc.) can consume snapshots programmatically.

Scope

• Headless CLI (memviz parse <pickle>, memviz analyze <dir>) that reuses the same Rust/WASM parser.
• Stable JSON output for downstream tooling.
• Agent skills: structured invocations that let a code agent ask focused questions ("top 10 leak suspects", "peak-by-rank", "compare rank N vs M").
• Document the contract (schema, error codes) so skills don't break with every internal refactor.

Notes

The WASM module already runs under Node (see bench/memviz.mjs), so the CLI can wrap it directly without a new parser implementation.

Tracked in: #5