feat: consent code compatibility model for search API

[NoopDog]

Summary

Add a consent code compatibility model to the search API so researchers can find datasets by describing their use case ("I'm studying diabetes", "I'm a for-profit company") rather than knowing exact consent code strings.

PRD: docs/PRD-consent-code-compatibility.md

Problem

The search API treats consent codes as opaque strings with exact matching. Three failures:

1. No permission hierarchy: "I'm studying diabetes" resolves to DS-DIAB (~9 studies), missing ~1,500 GRU and ~700 HMB studies also available for diabetes research
2. No base code expansion: "GRU" exact-matches 1,081 studies, missing 462 GRU-IRB/GRU-NPU/etc. variants
3. No modifier semantics: "I'm a for-profit company" has no mapping — the system doesn't understand that NPU means non-profit only

Solution

• Deterministic consent_logic.py module that parses codes (base + disease + modifiers), knows the permission hierarchy (GRU ⊇ HMB ⊇ DS-X), and computes eligible codes given a researcher profile
• New resolve agent tool compute_consent_eligibility(purpose, disease, is_nonprofit, explicit_code) returns all compatible codes
• Extract agent update to recognize eligibility language ("what can I use", "for-profit", "consented for") and generate consent mentions — consent filtering only activates when asked, not on every disease query
• Base code expansion so "GRU" returns all GRU-* variants; "GRU-IRB" stays exact

Acceptance Criteria

• consent_logic.py created with parse_consent_code, expand_disease, compute_eligible_codes + unit tests
• consent_codes.json updated with disease_hierarchy mapping
• Resolve agent has compute_consent_eligibility tool
• RESOLVE_PROMPT.md updated with eligibility-based resolution (Pattern A: explicit code, Pattern B: eligibility)
• EXTRACT_PROMPT.md updated with researcher profile extraction patterns
• Eval cases added for extract and resolve agents
• "GRU" query returns all GRU-* variants (1,543 vs 1,081 today)
• "for-profit cancer research" excludes NPU-modified codes
• "eligible for diabetes research" returns GRU + HMB + DS-DIAB family
• "diabetes studies" (no eligibility cue) remains focus-only — no consent filter added

Files to Create/Modify

File	Action
backend/concept_search/consent_logic.py	Create
backend/concept_search/consent_codes.json	Modify — add disease_hierarchy
backend/concept_search/index.py	Modify — add compute_consent_eligibility
backend/concept_search/resolve_agent.py	Modify — add tool
backend/concept_search/RESOLVE_PROMPT.md	Modify — eligibility strategy
backend/concept_search/EXTRACT_PROMPT.md	Modify — researcher profile patterns
backend/concept_search/eval_extract.py	Modify — add test cases
backend/concept_search/eval_resolve.py	Modify — add test cases
backend/tests/test_consent_logic.py	Create