test(medgemma): add tests for Gemma chat template formatting

feat(medgemma): switch to text_generation with Gemma chat template

fix: correct logger variable and test mocks for enhanced search_direct

test: add tests for SearchAnchors new fields and enhanced MCP search

feat(gemini): update search anchors prompt for interventions and eligibility keywords

feat(mcp): enhance search with intervention and eligibility dimensions

feat(models): add interventions and eligibility_keywords to SearchAnchors

test: update tests for evidence-linked mock data and new features

feat: add Parlant configurable log level

feat: add Gemini rate limiting and structured logging

feat: improve MedGemma retry logic and criterion evaluation

feat: add direct ClinicalTrials.gov API v2 search methods

feat: wire real API pipeline with demo mode and Gemini truncation fix

fix: update MCP test mocks for streaming HTTP transport

style: apply ruff format to entire codebase

feat: add ty type checker and fix async/type correctness

fix: make GeminiPlanner truly async with asyncio.to_thread

fix: cache service instances in Parlant tools to avoid per-call overhead

chore: fix lint issues across agent and test files

feat: add orchestrator entry point

feat: implement 5-state Parlant journey

feat: configure Parlant guidelines

feat: implement 7 Parlant tools

feat: implement Parlant agent setup

feat: add trial normalization and retry logic to MCP client

feat: complete Gemini planner with all 6 methods

feat: wire MedGemma extractor to HF Inference Endpoint

feat: add config module and update dependencies

Initial PoC codebase

feat: implement 3 BE service stubs with 17 TDD tests

feat: implement 5 Pydantic v2 data models with 37 TDD tests

chore: initialize project skeleton with pyproject.toml