feat: Add people and relationship memory

[QueryPlanner]

Priority: Medium

Level of Effort: Large (5-8 days)

Summary

Add a people/relationship memory system that lets the agent remember information about people mentioned by the user. This goes beyond generic memory — it should track personal details, relationship context, and conversation notes about specific individuals.

Scope

• New module: src/blacki/people/
• Storage: PostgresPeopleStorage (table: people, extends PostgresStorage base)
• Data model:
  • name (str, required)
  • relationship (str, optional — e.g., "coworker", "friend", "family")
  • context / notes (str, JSONB — flexible structured notes)
  • tags (list[str] — for categorization)
  • first_mentioned (datetime, auto-set)
  • last_updated (datetime, auto-updated)
• Tools:
  1. remember_person — add or update a person record with details
  2. search_people — semantic search or keyword search across stored people
  3. get_person — retrieve full details for a specific person by name
  4. list_people — list all known people, optionally filtered by relationship/tag
  5. delete_person — remove a person record
• Vector integration (optional/phase 2): Use the existing Mem0/Qdrant vector store for semantic search across people notes
• Pattern: Follow existing storage + tools patterns from calories/ and workouts/

Sources

• Existing memory tools: src/blacki/memory/tools.py
• Storage base class: src/blacki/storage/base.py
• Calorie storage pattern: src/blacki/calories/storage.py
• Calorie tools pattern: src/blacki/calories/tools.py
• Workout storage: src/blacki/workouts/storage.py
• Tool registry: src/blacki/registry.py
• DI container: src/blacki/container.py
• Memory config (vector store setup): src/blacki/memory/config.py
• AGENTS.md tool design principles

Passing Criteria

• uv run ruff check passes with no errors
• uv run ruff format passes with no changes needed
• uv run mypy . passes with no errors
• uv run pytest tests/people/ --cov=src/blacki/people passes with 100% coverage on people module
• uv run pytest full suite passes
• remember_person correctly stores and updates people records scoped to user_id
• search_people returns relevant results by name or keyword
• get_person returns full record including all stored fields
• list_people supports filtering by relationship type and tags
• delete_person removes the record and returns confirmation
• Storage is properly initialized via AppContainer and registry.py
• All tools gracefully handle missing/empty data with clear error messages