Clean up agent-docs: remove research/design records, keep it pure how-to-use guidance

[vivek7405]

Problem

agent-docs/ exists to give AI agents detailed guidance on how to USE webjs correctly (the topic docs loaded on demand: components, styling, built-ins, advanced, testing, etc.). It is NOT a place for research write-ups or design/decision records. Those have leaked in and dilute the directory's purpose: an agent loading agent-docs for how-to-use guidance should not wade through internal design history.

Known pollutant: agent-docs/ssr-partial-nav-design.md is a design record, present in the directory but NOT referenced in the root AGENTS.md detail-docs table (the table lists advanced, built-ins, components, configuration, deployment, framework-dev, lit-muscle-memory-gotchas, metadata, recipes, service-worker, styling, testing, typescript). Anything in agent-docs/ that is a research/design record rather than usage guidance should be relocated out and removed from the directory.

The convention going forward (set by the owner): a research/design investigation is recorded as a PR (good title, thorough description, deep-dive comments) and then CLOSED to keep the record, never as a file under agent-docs/.

Design / approach

1. Audit every file in agent-docs/ against one test: is this pure how-to-use-webjs guidance for an agent? If it is a research/design/decision record, it does not belong.
2. For each pollutant (starting with ssr-partial-nav-design.md): preserve the content as a closed PR record (or an issue comment if more appropriate), then delete the file from agent-docs/.
3. Reconcile the AGENTS.md detail-docs table with the actual directory so the two match exactly (every listed file exists, every existing file is listed). Note a likely inverse drift to verify: AGENTS.md references agent-docs/deployment.md but it was not present in the directory listing during the Research whether Drizzle is a better default ORM than Prisma for webjs apps #532 work, so the audit should catch BOTH unreferenced files (pollutants) and referenced-but-missing files (drift).

Implementation notes (for the implementing agent)

• Audit target: agent-docs/*.md. Compare against the detail-docs table in the root AGENTS.md (the markdown table near the top mapping File -> Topic).
• Known pollutant to remove: agent-docs/ssr-partial-nav-design.md (design record, unreferenced). Preserve its content first (closed-PR record per the new convention) before deleting.
• Verify the inverse: grep AGENTS.md for every agent-docs/<name>.md it references and confirm each file exists; fix any referenced-but-missing entry (e.g. deployment.md).
• Landmine: do NOT delete a file that IS legitimate usage guidance just because it is unreferenced; the test is content (usage guidance vs research/design), and the fix for a legit-but-unlisted doc is to ADD it to the AGENTS.md table, not delete it.
• The MCP knowledge layer bundles the agent-docs/*.md corpus (packages/mcp), so removing/renaming a file flows to the MCP automatically at prepack; confirm nothing in packages/mcp hardcodes a removed filename.
• Invariant: prose changes must respect AGENTS.md invariant 11 (no em-dashes, no pause hyphen/semicolon, no colon on a code-shaped LHS), enforced by .claude/hooks/block-prose-punctuation.sh.

Acceptance criteria

• agent-docs/ contains only how-to-use-webjs guidance; no research/design/decision records remain.
• ssr-partial-nav-design.md content is preserved as a closed-PR record and the file is removed from agent-docs/.
• The root AGENTS.md detail-docs table and the actual agent-docs/ directory match exactly (no unreferenced files, no referenced-but-missing files).
• No packages/mcp reference points at a removed/renamed agent-docs filename.