feat: extract and persist Findings from ToolResults (FactStore Phase 2)

[luceinaltis]

Summary

Closes #157.

Phase 2 of FactStore: the findings table sketched in Phase 1 is now populated. Discrete Finding records are extracted from every successful ToolResult.data dict in ConversationEngine._persist_facts and persisted alongside theses, so ticker-based recall (e.g. "what did we learn about AAPL's earnings last time?") no longer depends on free-text search.

What changed

• qracer/memory/fact_store.py — new findings table + idx_findings_entity index, plus save_finding() / get_findings() CRUD. Confidence is clamped to [0.0, 1.0] at the store boundary so noisy extractor inputs don't corrupt the schema.
• qracer/memory/finding_extractor.py — new module. extract_findings(ToolResult) -> list[FindingDraft] with dedicated zero-LLM-cost extractors for trade_thesis, news, and fundamentals. Unknown tools, failed results, and extractor exceptions all yield [] so a bad payload can never break the persistence pipeline.
• qracer/conversation/engine.py — _persist_facts now iterates over every analysis.results entry and persists each extracted draft; per-draft failures are isolated so one bad tool doesn't prevent siblings being saved.

Non-goals (per the issue)

• No LLM-based extraction — structured-data parsing only.
• UserPreference extraction is out-of-scope (Phase 3).

Test plan

• uv run pytest tests/memory/test_fact_store.py tests/memory/test_finding_extractor.py — 35 tests covering Finding CRUD (save, clamp, filter, order, limit, event_date) and extractor behaviour (failure paths, sentiment-weighted confidence, article cap, partial fundamentals, missing-ticker guards).
• uv run pytest full suite — 798 passed, 14 skipped.
• uv run ruff check — clean.
• uv run pyright qracer/memory/finding_extractor.py qracer/memory/fact_store.py qracer/conversation/engine.py — 0 errors.

Manual verification

from qracer.memory.fact_store import FactStore
store = FactStore()  # in-memory
store.save_finding(entity="AAPL", statement="P/E 29", confidence=0.9,
                   source_tool="fundamentals", session_id="demo")
store.get_findings("AAPL")  # -> [Finding(entity='AAPL', ...)]