fix: Ship the RAG index with the deploy -- it was never reaching the Space

data/ is (correctly) gitignored to keep multi-GB source PDFs and raw corpora
out of the repo, but that also caught the built index itself, so the Space
never had it: confirmed live via /api/status returning indexed_chunks: 0
right after deploy. The built index (chunks.json + bm25.pkl) is ~23MB, not
multi-GB, so it ships fine as a normal tracked file -- carved out an
exception in .gitignore for just those two files. Tracked both via Git LFS
since chunks.json exceeds the Hub's 10MB plain-blob limit.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

.gitattributes

@@ -0,0 +1,2 @@
+data/rag_index/chunks.json filter=lfs diff=lfs merge=lfs -text
+data/rag_index/bm25.pkl filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -14,11 +14,20 @@ __pycache__/
 ._*
 
 # Downloaded and generated data, books, and large corpora
-data/
+data/*
 *.pdf
 *.djvu
 *.epub
 
+# ...except the built RAG index itself (~23MB, not the multi-GB source PDFs
+# it was built from) -- this must ship with every deploy or the Space serves
+# zero retrieved chunks (observed directly: /api/status reported
+# indexed_chunks: 0 with this excluded, since data/ is never pushed to the
+# Space's git repo).
+!data/rag_index/
+!data/rag_index/chunks.json
+!data/rag_index/bm25.pkl
+
 # Model checkpoints and fine-tuned weights
 adapters/
 checkpoints/

data/rag_index/bm25.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a44f807feb5042e172378e8a08783266ea1a6aa5788fc37aa5adb28bf2fc20db
+size 9492608

data/rag_index/chunks.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b744f1fb462b3cfc345f24411853b88b6b23fda26459632990a8fc4d3b24841
+size 12949232