flowTorch 1.8.0

This release adds scalable statistics and modal-analysis workflows for snapshot datasets that exceed the memory of a single process or node.

Highlights

Add unified batched snapshot statistics for the first four moments, data-fraction dependency, spatial-field difference norms, linear trends, spatial min/max/mean and selected quantiles, and 50-bin spatiotemporal histograms.
Add optional boolean spatial masks and distributed statistics across CPU nodes or GPUs behind the same public interfaces used for serial execution.
Add lazy multi-field state vectors with component-major packing and one normalization factor per physical field, including weighted RMS fitting.
Add shared-memory and spatially distributed out-of-core TSQR SVD for extremely tall, skinny data matrices.
Add incremental SVD updates as new snapshots arrive and versioned SVD save/load checkpoints that preserve lazy source-backed execution.
Extend PAMSPOD to consume lazy sources or reuse an existing serial or distributed SVD.
Keep SVD modes, SPOD modes, mode reconstructions, and partial reconstructions lazy and allow splitting them back into their original vector, tensor, and scalar fields.
Add direct spatial slicing for HDF5 and FOAM NumPy dataloaders.
Add distributed SVD/SPOD, Slurm hybrid-execution, batching, update, reconstruction, and checkpoint documentation.
Adopt develop as the integration branch for contributions and refresh the README to feature SPOD workflows.