Data I/O contract plan

Scope

Keep this branch focused on dataset layout and HDF5/NPY boundaries in dfode_kit/data_operations.

Audit summary

• dfode_kit/data_operations/h5_kit.py currently mixes three concerns: HDF5 inspection/loading, reactor/model integration, and error reporting.
• HDF5 layout assumptions are implicit: code assumes a root mechanism attribute and a scalar_fields group of stack-compatible datasets.
• dfode_kit/cli/commands/h52npy.py reimplements part of the HDF5 read path instead of sharing one validated loader.
• Some runtime contracts still rely on loose assumptions or non-deterministic dataset iteration.

First slice

1. Extract lightweight HDF5 contract helpers into a focused module.
2. Define and enforce the scalar_fields layout contract:
3. required group name
4. deterministic dataset ordering
5. each dataset must be a 2D array
6. all datasets must share the same column count
7. required root attrs can be checked explicitly where needed
8. Reuse those helpers from h5_kit.py and the h52npy CLI command.
9. Add fast tests for the contract helpers.

Follow-up slices

• Split pure HDF5 I/O from Cantera/Torch integration paths in h5_kit.py.
• Add NPY state-matrix contract helpers shared by labeling/augmentation commands.
• Add lightweight CLI smoke tests for h52npy and label contract failures.