feat: RCP EPFL model

[caviri]

Note

Integrates EPFL RCP (OpenAI‑compatible) model across all agents, fixes/implements token estimation and usage tracking, and hardens academic-catalog enrichment, Infoscience tools, and JSON‑LD API handling.

• LLM/Config:
  • Add openai-compatible EPFL RCP model to run_llm_analysis, run_user_enrichment, run_organization_enrichment, run_academic_catalog_enrichment, and run_epfl_assessment.
  • Implement OpenAIProvider with base_url in create_pydantic_ai_model; add RCP_TOKEN in .env.example.
• Token Usage Tracking:
  • Implement client-side token estimation and proper usage extraction in epfl_assessment; fix estimated token keys in academic_catalog_enrichment.
  • Propagate usage aggregation/logging across repository/user/org analysis and enrichment flows.
• Academic Catalog Enrichment:
  • Use structured AcademicCatalogEnrichmentResult (repo/author/org maps) with direct assignment; update prompts and repository/user/org runners.
  • Infoscience client/tooling improvements (person/orgunit search, profile_url fix, caching) and robust parsing.
• API/Models/JSON‑LD:
  • Preserve raw JSON‑LD via APIOutput union/serializer; add validation/logging in endpoints.
  • Typing and model cleanups (list annotations), minor cache config and logging tweaks.

^{Written by Cursor Bugbot for commit dd5c611. This will update automatically on new commits. Configure here.}