feat: add Variables tab to study detail page

[NoopDog]

Summary

Add a "Variables" tab to the study detail page that shows the variables measured in each dbGaP study, organized by concept category.

Requirements

1. Category Tree Display

Variables are grouped into a tree using the existing classification hierarchy:

NCPI Category (e.g. "ncpi:biomarkers")
  └─ Concept (e.g. "topmed:hdl_cholesterol")
       └─ Variable (e.g. "HDL - HDL cholesterol mg/dL")  ← only for small studies

• Only show populated categories — omit any NCPI category or concept with zero variables in the study.
• Walk the ISA hierarchy (concept-isa.json) from each variable's concept_id up to its NCPI category (ncpi-categories.json).
• Unclassified variables (no concept_id) go under an "Other / Unclassified" group.

2. Tiered Display by Study Size

• ≤200 variables: Show the full tree including individual variables (name + description) under each concept.
• >200 variables: Show the tree down to the concept level only (category → concept with variable counts). Individual variables are not listed.

Study size distribution: 2,721 studies (95%) have ≤200 variables; 143 studies have >200.

3. Data Sources (JSON only, no DuckDB needed)

All data can be computed at build time from these files:

File	Purpose
catalog-build/classification/output/llm-concepts-v4/phs*.json	Variables with concept_id per study
catalog-build/classification/output/concept-isa.json	Concept → parent category mapping
catalog-build/classification/output/ncpi-categories.json	Top-level NCPI category names/descriptions
catalog-build/classification/output/concept-vocabulary.json	Concept names and descriptions

4. Build-Time Computation

For each study:

1. Read llm-concepts-v4/phs{study}.json
2. Group variables by concept_id
3. Walk ISA hierarchy to resolve each concept to its NCPI category
4. Emit a tree structure: { category → { concept → [variables] } }
5. Include variable count per concept (always) and variable details (only if study total ≤200)

Design

• Add a new "Variables" tab on the study detail page, next to the existing Publications tab.
• Tree is collapsible by category.
• Support search/filter within the variable list for studies that show individual variables.