docs: make agent context retrieval-oriented (#70)

[jlapenna]

Summary

Replace sparkstack's oversized procedural agent context with a retrieval-first router and explicit ownership map.

Changes

• reduce AGENTS.md from 8.6 KB to 3.0 KB and route domain work to existing skills
• add architecture and documentation-discovery maps
• add six executable context contracts to the unit suite
• preserve registry and sibling-source boundaries without duplicating their procedures

Verification

• focused context contracts: 6 passed
• non-live unit/regression suite: 139 passed, 1 failed
• pre-push checks excluding uv-lock: Ruff, formatting, Pyright, gitleaks, and Bandit passed
• git diff --check

The one non-live failure is environmental: the local sibling checkout provides sparkrun 0.2.32 while this repository's committed lock expects 0.2.36, changing the builder's effective context calculation (3481 vs 4096). The same mismatch makes the uv-lock hook request a lock rewrite. No lockfile change is included; CI should validate against its clean dependency checkout.

No live E2E was run because this is documentation/source-contract-only work and the repository requires a deliberately prepared stack and sibling branch prerequisites before E2E.

Operational impact

No stack update, model rotation, service mutation, registry change, or deployment.

Closes #70

Provenance

Codex thread: 01a0faae-74de-7110-a2b8-d326048e984a

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.