feat: surface memex PersonalContext in orchestrator system prompt

[toadkicker]

Summary

• Introduces PersonalContext — a compact JSON-LD preamble built from EpisodeDB history, AgentProfile aggregates, and the user's TraitBeacon Person document
• Injects context into the orchestrator's LLM calls via a tokio::sync::watch channel; consumers borrow the latest preamble lazily at call time (zero polling overhead)
• Scope: OnDeviceAiExecutor + build_orchestrator_prompt_with_template only. pap-agents / DynamicAgentHandler are untouched.
• New Tauri command save_trait_beacon_profile lets the frontend persist the user's Schema.org Person document; triggers an immediate preamble refresh
• 365 tests pass (241 papillon-shared, 124 papillon); 10 new tests added

Key files

• crates/papillon-shared/src/personal_context.rs (new)
• apps/papillon/src/state.rs — context_tx watch sender + trait_beacon_profile Arc
• apps/papillon/src/agents/on_device_ai.rs — lazy preamble inject
• apps/papillon/src/inference.rs — context: Option<&str> param
• apps/papillon/src/commands/episodes.rs / orchestrator.rs — push after write
• apps/papillon/src/commands/agents.rs — save_trait_beacon_profile command

Test plan

• All CI checks green (13 jobs)
• Smoke test: issue On-Device AI query after recording an episode, confirm [PAP PERSONAL CONTEXT] block appears in RUST_LOG=debug output
• save_trait_beacon_profile with @type: "Person" persists and refreshes preamble
• Empty DB path: no preamble injected (no noise in first-run experience)

🤖 Generated with Claude Code

[github-actions]

Benchmark Regression Report

PAP Protocol Benchmark Regression Check
========================================
Baseline: .bench-baseline/baseline.json
Threshold: 30%

  ed25519_keypair_generation                19.7 µs  (baseline: 19.7 µs, +0.1%)  [ok]
  did_key_derivation                         1.5 µs  (baseline: 1.5 µs, -1.7%)  [ok]
  mandate_create_sign                       24.0 µs  (baseline: 24.0 µs, -0.2%)  [ok]
  mandate_chain_verify_depth3              130.5 µs  (baseline: 139.0 µs, -6.1%)  [ok]
  sd_jwt_issue_5claims                      27.9 µs  (baseline: 27.9 µs, +0.0%)  [ok]
  sd_jwt_verify_disclose_3of5               46.1 µs  (baseline: 48.6 µs, -5.1%)  [ok]
  session_open_full_lifecycle              112.3 µs  (baseline: 108.6 µs, +3.3%)  [ok]
  receipt_create_cosign                     47.9 µs  (baseline: 47.9 µs, +0.1%)  [ok]
  federation_announce_local                 55.8 µs  (baseline: 61.3 µs, -9.0%)  [ok]

All benchmarks within 30% of baseline.

_{Threshold: 10% regression vs baseline from main}