feat(agent-ready): add session-startup, DoD, and JSON-ledger guidance

[dgalarza]

Summary

Adds three patterns to the AGENTS.md template drawn from Anthropic's harness design articles (Effective harnesses for long-running agents, Harness design for long-running application development), addressing failure modes the existing scaffold did not cover:

• Session Startup — bearing-getting ritual (pwd, git log, smoke test) agents run on every fresh context. Prevents the "look around, declare done" failure mode where an agent piles new work on top of an existing broken state.
• Definition of Done — codifies end-to-end verification before marking work complete. Targets the most common over-confidence failure: agents shipping changes because "unit tests pass" without exercising the feature end-to-end.
• JSON over Markdown for ledgers — short note in Template Notes capturing Anthropic's finding that agents are far less likely to inappropriately edit JSON files than Markdown ones. Useful guidance when a project adds a task ledger or feature-status file.

agents-md mode populates Session Startup and DoD during generation using detected build/test commands. audit mode now checks for both sections, flags DoD sections that omit end-to-end verification, and offers matching auto-fixes.

Bumps agent-ready to 1.3.0 in plugin.json and marketplace.json.

Test plan

• Run audit mode against a project missing Session Startup — confirm it flags as missing and offers fix
• Run audit mode against a project with DoD but no end-to-end mention — confirm "present but does not mention end-to-end verification" warning fires
• Run agents-md mode on a fresh project — confirm generated AGENTS.md contains both new sections with detected commands filled in
• CI: skill validator + JSON validation pass