fix(hooks): restrict prompt entity injection types

[NicholaiVogel]

Summary

• restrict prompt-submit entity injection to person and project entities
• keep possessive normalization and longest-span selection while filtering generic one-token entity names
• mirror the selector behavior in daemon-rs and add TS/Rust regressions for disallowed entity types

Validation

• bunx biome check platform/daemon/src/hooks.ts platform/daemon/src/hooks.prompt-submit.test.ts
• bun test platform/daemon/src/hooks.prompt-submit.test.ts
• cargo test -p signet-daemon prompt_submit
• bun run build:signetai
• signet daemon restart --no-sync
• live hook matrix: signet prompt context and signets prompt context inject; projects roadmap, agents are useful, tools setup, claude code connector setup, and context stay silent

Notes

• No database migrations.
• No public API contract change; this narrows prompt-submit entity selection policy.
• nicholais favorite pens currently resolves the person entity but returns no-aspect-hit because the clean pen fact is not represented as a current Nicholai ontology attribute.

PR Readiness (MANDATORY)

• Spec alignment validated (INDEX.md + dependencies.yaml)
• Agent scoping verified on all new/changed data queries
• Input/config validation and bounds checks added
• Error handling and fallback paths tested (no silent swallow)
• Security checks applied to admin/mutation endpoints
• Docs updated for API/spec/status changes
• Regression tests added for each bug fix
• Lint/typecheck/tests pass locally

Migration Notes (if applicable)

• Migration is idempotent
• Daemon Rust parity reviewed or explicitly N/A
• Rollback / compatibility note included in PR description

[PR-Reviewer-Ant]

Review posted through GitHub's PR review UI for commit 15867931.

[PR-Reviewer-Ant]

Review metadata

• Reviewer: pr-reviewer
• Model: gpt-5.5
• Commit: 15867931

I did not find a concrete blocker in this diff. The implementation matches the PR description: prompt-submit entity selection is narrowed to person/project entities, generic one-token entities are filtered, possessive matching is preserved, and the Rust shadow path has matching regressions.

Confidence: Medium [sufficient_diff_evidence, missing_runtime_repro] - The selector behavior is directly visible in platform/daemon/src/hooks.ts and mirrored in platform/daemon-rs/crates/signet-daemon/src/routes/hooks.rs, with targeted TS and Rust tests covering possessives, disallowed entity types, generic plural terms, and longest non-overlapping spans. I did not run the stated validation commands or inspect the live ontology data distribution, so residual risk is mainly around real-world scoring thresholds rather than a provable diff bug.