Studio: move RAG deps to dedicated rag.txt and install in normal path

no-torch-runtime.txt is only consumed in NO_TORCH (Intel Mac
GGUF-only) mode, so qdrant-client / bm25s / pymupdf etc. were never
installed in the normal install path. Split them into rag.txt and
add a step to install_python_stack.py that installs it after studio
deps, skipped only when NO_TORCH is set.

studio/backend/requirements/no-torch-runtime.txt

@@ -76,24 +76,3 @@ trl>=0.18.2,!=0.19.0,<=0.24.0
 sentence-transformers
 cut_cross_entropy
 pillow
-
-# RAG: vector store, lexical index, document parsers.
-# qdrant-client supports a pure-Python local mode (QdrantClient(path=...))
-# that we use to keep the studio install self-contained — no separate
-# server. bm25s persists per-scope indices to disk.
-qdrant-client>=1.12
-bm25s>=0.2
-# RAG parsers (Phase 3A): layout-aware Markdown extraction so the chunker
-# can split on real headings instead of running paragraphs together.
-# pymupdf4llm preserves headings + pipe-tables; mammoth handles DOCX
-# Heading styles; markdownify converts HTML <h*>/<table>/<ul> faithfully.
-pymupdf>=1.24
-pymupdf4llm>=0.0.17
-mammoth>=1.7
-markdownify>=0.13
-# pypdf is kept as a fallback for malformed PDFs that defeat pymupdf.
-pypdf>=4.0
-python-docx>=1.1
-beautifulsoup4>=4.12
-lxml>=5.0
-chardet>=5.2

studio/backend/requirements/rag.txt

@@ -0,0 +1,27 @@
+# Studio RAG dependencies.
+# Installed by studio/install_python_stack.py in the normal (with-torch)
+# path. Skipped in NO_TORCH (Intel Mac GGUF-only) mode because RAG
+# embedding relies on sentence-transformers, which requires torch.
+
+# Vector store + lexical index. qdrant-client supports a pure-Python
+# local mode (QdrantClient(path=...)) that we use to keep the studio
+# install self-contained — no separate server. bm25s persists per-scope
+# indices to disk.
+qdrant-client>=1.12
+bm25s>=0.2
+
+# Layout-aware Markdown extraction (Phase 3A) so the chunker can split
+# on real headings instead of running paragraphs together. pymupdf4llm
+# preserves headings + pipe-tables; mammoth handles DOCX Heading styles;
+# markdownify converts HTML <h*>/<table>/<ul> faithfully.
+pymupdf>=1.24
+pymupdf4llm>=0.0.17
+mammoth>=1.7
+markdownify>=0.13
+
+# pypdf is kept as a fallback for malformed PDFs that defeat pymupdf.
+pypdf>=4.0
+python-docx>=1.1
+beautifulsoup4>=4.12
+lxml>=5.0
+chardet>=5.2

studio/install_python_stack.py

@@ -915,6 +915,8 @@ def install_python_stack() -> int:
         base_total -= 1  # triton step is skipped on macOS
     if not IS_WINDOWS and not IS_MACOS and not NO_TORCH:
         base_total += 3
+    if not NO_TORCH:
+        base_total += 1  # studio RAG deps (rag.txt)
     _TOTAL = (base_total - 1) if skip_base else base_total
 
     # 1. Try to use uv for faster installs (must happen before pip upgrade
@@ -1203,6 +1205,17 @@ def install_python_stack() -> int:
         req = REQ_ROOT / "studio.txt",
     )
 
+    # 8b. RAG dependencies (vector store + lexical index + layout-aware
+    #     parsers). Skipped in no-torch mode because RAG embeddings go
+    #     through sentence-transformers, which requires torch.
+    if not NO_TORCH:
+        _progress("rag deps")
+        pip_install(
+            "Installing RAG dependencies",
+            "--no-cache-dir",
+            req = REQ_ROOT / "rag.txt",
+        )
+
     # 9. Data-designer dependencies
     _progress("data designer deps")
     pip_install(