Add MCP Lens Schema Audit to agent-specific evaluation

[labmimors]

Why this belongs

This adds one deterministic GitHub Action to §9 · Agent-specific evaluation. Given a checked-in model-facing MCP tool payload, it reports tool count and Tool Schema JSON UTF-8 bytes and can fail CI on explicit max-tools or max-schema-bytes thresholds.

This is deliberately a component-level check for the tool surface presented to an agent, not an end-to-end task or agent-quality evaluation.

Show your work

• The Action contract fixes the Node 24 runtime, inputs, outputs, and budget behavior.
• The implementation uses no model call, API key, or network request.
• The public Action tests cover accepted payload shapes, numeric outputs, threshold failures, and Step Summary behavior. I reproduced all 9 tests in a fresh checkout before opening this PR.

Verification

• Confirmed both canonical links return HTTP 200.
• Checked this repository plus open/closed GitHub issues and PRs for dsh-mcp-lens, MCP Lens, and the canonical repository URL; no existing submission was found.
• Ran git diff --check; the content change is one README entry.

Disclosure

I am the maintainer of labmimors/dsh-mcp-lens.