perf: switch research agent from Opus to Haiku (#21)

RESEARCH_MODEL was claude-opus-4-7 (~$15/MTok input). Switching to
claude-haiku-4-5-20251001 (~$0.80/MTok input) cuts the per-session
research cost by ~19x. The research task (call search tool, synthesize
trial results) does not require Opus-level reasoning.

Estimated saving: ~$0.22/session (~70% of total session cost).

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

config.py

@@ -1,4 +1,4 @@
 CTGOV_BASE = "https://clinicaltrials.gov/api/v2/studies"
 INTAKE_MODEL = "claude-sonnet-4-6"
-RESEARCH_MODEL = "claude-opus-4-7"
+RESEARCH_MODEL = "claude-haiku-4-5-20251001"
 ELIGIBILITY_MODEL = "claude-sonnet-4-6"