feat: citations, eval sidecar/pvalue/trace, nomic prefixes

mcp 2.x already on main. Does not replace the variety chunker
(one chunk per variety is the anti-hallucination contract).

- Numbered [1] citations on search_docs / search_trials
- Eval JSONL sidecar, k-curve, eval.pvalue, eval.trace
- Nomic prefixes at embed time only; stored text unprefixed

Closes #25

rag/index.py

@@ -23,7 +23,7 @@ import chromadb
 from chromadb.config import Settings
 
 from .chunk import chunks_from_variety, chunks_from_trial
-from .embeddings import embedding_function
+from .embeddings import EMBED_DOC_PREFIX, embed_texts, embedding_function
 
 log = logging.getLogger(__name__)
 logging.basicConfig(level=logging.INFO, format="%(asctime)s  %(message)s")
@@ -83,9 +83,12 @@ def upsert_to_chroma(records: list[dict]) -> int:
     total = 0
     for i in range(0, len(records), BATCH):
         chunk = records[i:i + BATCH]
+        texts = [r["text"] for r in chunk]
+        vectors = embed_texts(texts, prefix=EMBED_DOC_PREFIX, ef=embedding_function())
         col.upsert(
             ids=[r["id"] for r in chunk],
-            documents=[r["text"] for r in chunk],
+            documents=texts,
+            embeddings=vectors,
             metadatas=[r["metadata"] for r in chunk],
         )
         total += len(chunk)