Fast-subsub

[sirbastiano]

This PR improves subaperture processing in two areas:

1. Subaperture pipeline refactor (sarpyx/processor/core/subaperture_full_img.py)

• Adds automatic backend selection between NumPy and Dask based on memory pressure.
• Introduces chunked, lazy Dask loading for I/Q ENVI pairs.
• Adds helper utilities for image shape probing and memory estimation.
• Converts complex image creation paths to favor complex64 for reduced memory usage.
• Adds backend-aware sublook generation and safer recentering in Dask path.
• Adds/updates vectorized helper behavior (DeHammWin) consistency for performance and dtype stability.

2. Robustness fixes in Zarr metadata handling (sarpyx/utils/zarr_utils.py)

• Adds recursive, attribute-safe serializers for metadata/ephemeris values.
• Handles non-JSON-safe types (NumPy scalars, pd.Timestamp, Enum, NaN, Path, bytes, etc.).
• Prevents broken Zarr attributes for complex/pandas/dtype-heavy metadata payloads.

3. Test coverage expansion (tests/test_subaperture_dask.py)

• Adds tests for:
  • NumPy vs Dask numerical equivalence.
  • NumPy and Dask output shapes/dtypes.
  • Output ENVI file creation and shape checks.
  • Backend auto-detection and force-Dask override behavior.
  • Dask recenter/roll behavior and complex64 dtype consistency.
  • Vectorized DeHammWin behavior (1D/2D).

Related commits

• f8ae932 — enhance complex image handling and add comprehensive Dask subaperture tests
• 2724054 — subfix + Zarr utility serialization fixes

Files changed

• sarpyx/processor/core/subaperture_full_img.py
• sarpyx/processor/core/utilis.py
• sarpyx/utils/zarr_utils.py
• tests/test_subaperture_dask.py

Notes

• Source branch: fast-subsub
• Target branch: main