Add batch upload, model selector, PDF export, and charts

[pypi-ahmad]

Summary

This PR implements the four upgrades listed in the project's README roadmap and keeps the app focused on TabFM + Ollama:

1. Batch Upload (one train CSV + multiple test CSVs)
2. Ollama model selector dropdown (auto-discovered from local /api/tags)
3. PDF export for insights
4. Plotly charts for prediction/confidence metrics

Key Changes

• Added CSV Upload and Batch Upload modes in Streamlit UI.
• Added local Ollama model discovery and selectable model for insight generation.
• Added downloadable PDF reports for single-run and batch-run insights.
• Added Plotly charts:
  • predicted class distribution
  • confidence distribution (single mode)
  • aggregate class distribution (batch mode)
• Improved prediction flow to avoid duplicated inference calls in-app by deriving labels from one probability pass.

Dependency Updates

• Added:
  • plotly
  • reportlab
  • safetensors (for environments where TabFM load requires it)

Validation

Ran locally:

• python3 -m py_compile app.py
• python3 -c "from tabfm import tabfm_v1_0_0_pytorch as b; b.load(); print('TabFM OK')"
• Manual Streamlit checks for:
  • single CSV mode
  • batch mode with multiple test CSVs
  • model selector behavior
  • PDF downloads
  • chart rendering

Notes

• outputs.md generated artifacts were intentionally excluded from this PR to keep the review diff clean.