Add low-rank waveform decomposition pipeline

[micatlan]

Summary

• add a native lowrank_waveform_decomposition pipeline that consumes the shared beat-aligned segment waveform state
• compute acquisition-wide SVD endpoints and independent per-beat SVD endpoints for raw and bandlimited arteries, with optional venous processing
• add pipeline discovery/settings, schema paths, HDF5 metadata, dependency scheduling, and focused tests

Why

The AngioEye low-rank waveform analysis needed a native EyeFlow implementation that reuses the existing velocity pipeline and follows the endpoint definitions in the svd_endpoints reference manuscript.

Endpoint definitions

• acquisition summaries aggregate all valid beat-location waveform columns
• R0 is the median centered-waveform RMS; TPR is retained only as a compatibility alias
• rho_m = R_m / R0, with rho0 = 1
• MPR is the median of paired local mean-to-pulsatile ratios
• alpha compares retained higher-mode energy with Mode 1
• G1 is the relative gap between the first two singular values
• beat-level outputs use an independent SVD for each beat

Validation

• python -m unittest test.test_lowrank_waveform_decomposition -v: 9 passed
• relevant regression tests: 22 passed
• ruff check --select E,F,I ...: passed
• full test discovery: 146/147 passed; the remaining failure is an unrelated Windows short-path versus long-path assertion in test_run_service.RunServiceTests.test_recursive_folder_expansion_is_sorted