Extend the agent self-API: conversation read/fork/colors, memory history, skill restore, viewer-scoped files/deployments, projects

[ReganBell]

Extends the agent self-API with seven capabilities agents kept needing and could not reach:

• Conversation colors — update a conversation's color
• Memory history & restore — list memory versions and restore one
• Skill restore — restore an archived skill
• Conversation read & fork — read a conversation the actor can see (tailTurns/beforeSeq paging) and fork one up to a sequence point
• Viewer-scoped file reads — fetch a file through the viewer's own visibility
• Projects — expose projects to agent capabilities
• Viewer-scoped deployment fetch — fetch a deployment the viewer can reach

Each endpoint is documented in the API catalog and covered by route tests (87 assertions across the eight touched test files).

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}