Fix kwarg timeframe to klines_tf in VPIN pre-scan

Fix: add numpy/pandas/gc imports inside fetch_vpin_cloud for Modal container

VPIN: Vectorized processing for extreme speed and heartbeat logs

VPIN: strictly sequential, robust ZIP handling, and NumPy fix

Fix VPIN: add numpy to image and handle empty zips in yield_vision_zips

Strictly sequential VPIN: process month-by-month with persistent state

Ultra-optimized VPIN: Klines for pre-scan + batch processing loop

Optimize Modal memory: sequential processing with disk-backed storage and explicit gc

Add VPIN Flow Toxicity feature

Sync Modal parallel architecture across entry points

Implement Auto-upload to Hugging Face from Modal Cloud

Route all data processing through Modal.com cloud (8GB RAM, 2h timeout)

Initial commit: React + FastAPI Dashboard migration