chore: pin deps + add bs4/html5lib/pdfminer for Spaces

requirements.txt

@@ -1,27 +1,34 @@
-# Core dependencies
-streamlit>=1.36
-pydantic>=2.7
-python-dotenv>=1.0
-pandas>=2.2
-numpy>=1.26
+# ---------- Core UI ----------
+streamlit==1.36.*
+pydantic==2.7.*
+python-dotenv==1.0.*
+pandas==2.2.*
+numpy==1.26.*          # safer with faiss-cpu wheels on Spaces
 
-# RAG/Search functionality (your existing)
-faiss-cpu>=1.8
-sentence-transformers>=3.0
-requests>=2.32
-pyyaml>=6.0
-tqdm>=4.66
+# ---------- RAG / Search ----------
+faiss-cpu==1.8.0
+sentence-transformers==3.0.*   # requires torch>=2; see below
+requests==2.32.*
+PyYAML==6.0.*
+tqdm==4.66.*
 
-# AWS Chronos for time series forecasting
-git+https://github.com/amazon-science/chronos-forecasting.git
-torch>=1.9.0
-transformers>=4.21.0
+# ---------- HTML / PDF parsing (adds bs4 fallbacks) ----------
+beautifulsoup4==4.12.*
+html5lib==1.1           # pure-Python parser fallback (no system libs)
+lxml==4.9.*             # optional: faster; our code falls back if missing
+pdfminer.six==20231228  # PDF text extraction
 
-# Financial data and visualization
-yfinance>=0.2.0
-plotly>=5.0.0
-matplotlib>=3.5.0
+# ---------- Optional ML stack (needed for s-transformers 3.x) ----------
+torch==2.2.*            # CPU wheels available on Spaces
+transformers==4.41.*    # modern, compatible with s-transformers 3.x
+accelerate==0.30.*      # optional, safe pin
 
-# Additional utilities
-python-dateutil>=2.8.0
-accelerate>=0.20.0
+# ---------- Financial / Viz (unchanged, pinned) ----------
+yfinance==0.2.*
+plotly==5.22.*          # any 5.x works; pin to recent
+matplotlib==3.8.*       # modern, but not bleeding-edge
+
+# ---------- (Optional) Chronos time-series ----------
+# NOTE: This adds heavy deps and longer build times.
+# Keep commented unless you actually use Chronos in the Space.
+# git+https://github.com/amazon-science/chronos-forecasting.git