Add governed chat agent discovery and recommendation

[paullizer]

Summary

• extend the Phase 8A inventory, proposal, decision, inline card, and durable resume contracts to governed unselected personal, global, and group agents
• add explicit discoverable_by_orchestrator policy and bounded safe descriptors, defaulting existing and new agents to undiscoverable
• reauthorize the exact source turn and rebuild the current governed catalog at decision, resume, and invocation before resolving an opaque scope-namespaced agent reference
• execute approved agents once as required discovery_approved evidence sources, disable inherited tools, and hand the terminal ledger to the central response finalizer
• add shared agent-policy authoring controls, safe inert card rendering, documentation, release notes, and version 0.250.067

Security And Governance

• catalog construction occurs only after current personal ownership, global feature/item policy, group membership/status, enabled/hidden, local-runtime, action-free, creation-identity, and descriptor checks
• proposal and browser surfaces contain only an opaque reference, display label, scope class, capability/evidence tags, read-only/external state, risk, sensitivity, latency, and cost
• instructions, secrets, endpoints, connector settings, canonical IDs, group details, attached actions/arguments, hidden tools, assigned-knowledge details, inaccessible entries, and unsafe counts stay outside planner/proposal/browser metadata
• descriptor or identity changes, deletion, disablement, policy loss, action attachment, runtime changes, or membership revocation invalidate approval
• discovered invocations suppress prompt/response previews and raw plugin citations, clear reusable agent state, disable function choice at agent/execution/service levels, and restore normal selected-agent behavior afterward

Validation

• orchestration, durable choice, planner/runtime, evidence, synthesis, and no-tools contracts: 105 passed
• agent schema/catalog/scope/Foundry compatibility: 44 passed, 2 unrelated stale source-text assertions deselected
• authenticated capability/agent decision and resume routes: 20 passed
• route Blueprint and unauthenticated policy contracts: 12 passed
• desktop/mobile choice-card Playwright: 5 passed
• authenticated desktop/mobile agent modal validation completed against the local app, including default-off, local-only, action-free, edit reconstruction, ARIA, keyboard, and overflow checks
• Python compilation, JavaScript syntax, JSON schema parsing, VS Code diagnostics, whitespace, full-file Broken Access Control (12 files), and XSS (16 files) checks passed

Deliberate Phase Boundaries

• no automatic discovered-agent invocation; explicit user choice remains required
• no second-agent recommendation when an agent is already selected
• Foundry-backed and action-attached agents remain explicitly selectable but undiscoverable in Phase 8B
• generalized multi-agent orchestration, consequential/write approval, and durable in-flight execution remain out of scope

Refs #1021