Add agent-backed PubMed fact verifier

[mindbomber]

Summary

• Adds an opt-in AgentFactVerifier that retrieves PubMed abstracts and asks the configured LLM caller for a structured JSON verdict.
• Keeps the existing keyword-backed FactVerifier as the fallback when no LLM caller is configured or the agent path fails.
• Adds FactCheckConfig.use_agent and wires MedGuard so the agent verifier receives the configured LLM caller.
• Adds regression tests for reasoned evidence, fallback behavior, and MedGuard component wiring.

Fixes #1.

Verification

• python -m pytest tests\test_fact_check.py tests\test_pipeline.py -q -> 16 passed
• python -m ruff check medguard\guardrails\fact_check.py medguard\knowledge\pubmed.py medguard\config.py medguard\core.py tests\test_fact_check.py -> passed
• python -m compileall medguard\guardrails\fact_check.py medguard\knowledge\pubmed.py medguard\config.py medguard\core.py tests\test_fact_check.py -> passed
• git diff --check -> clean apart from Windows LF/CRLF warnings

AANA gate

• Intake gate: pass, recommended action accept
• Code-change guardrail: pass, recommended action accept
• AIx score: 1.0
• Violations: none

Scope note

The agent verifier is a source-grounding aid, not a medical or compliance authority. It only scores claims against retrieved PubMed abstracts and preserves fallback behavior when evidence or LLM output is unavailable.