feat: add Pinecone skill for scientific RAG persistence layer

[immuhammadfurqan]

Summary

Adds a Pinecone skill positioned as the retrieval persistence layer that complements existing database-lookup, paper-lookup, and biopython skills. While those skills handle direct API access to public databases, Pinecone enables persistent semantic retrieval over embedded scientific data — necessary when repeated API queries are too slow, when working with proprietary research data, or when building multimodal RAG pipelines.

What's Included

• SKILL.md (403 lines) covering index management, batch upserting, namespaces, metadata filtering, hybrid search, and multimodal retrieval
• references/index_types.md — serverless vs pod tradeoffs, dimension selection table for 12 scientific embedding models
• references/scientific_embedding_models.md — domain to model mapping with usage patterns
• references/hybrid_search.md — dense + sparse BM25 setup with alpha-blending tuning guide
• scripts/index_pubmed.py — working PubMed abstract ingestion pipeline
• scripts/multimodal_radiology.py — working multimodal (text + image) ingestion pipeline

Scientific Workflows Covered

• PubMed literature RAG with voyage-large-2
• Multimodal radiology retrieval with voyage-multimodal-3
• Clinical note similarity with Bio_ClinicalBERT
• Molecular similarity search (Morgan FP via Pinecone for fast recall, RDKit Tanimoto for precise rerank)

Integration Points

Explicitly cross-references and integrates with: paper-lookup, database-lookup, biopython, pysam, scanpy, scientific-writing, pyhealth.

Scanner Results

• 0 CRITICAL, 0 HIGH findings
• 3 MEDIUM findings — all network-related (expected for any network-dependent skill; Pinecone is a cloud API by definition)

Validation

• Frontmatter validates as YAML matching the anndata/SKILL.md schema
• Both scripts pass python -m py_compile
• Cisco AI skill scanner: OK SAFE
• Tested with pinecone>=6.0.0
• Production-validated patterns from a multimodal medical chatbot (voyage-multimodal-3 + Pinecone)

Checklist

• Adheres to Agent Skills Specification
• SKILL.md with overview, when-to-use, capabilities, workflows, troubleshooting
• All cross-referenced files exist
• No hardcoded API keys (uses os.environ)
• References to official documentation included