HDF5 state format v0.2

[hexane360]

Adds new features to HDF5 state format, including probe and object type (in support of parameterized probe/object) as well as probe, object, and scan metadata. This change is backwards compatible (can open old HDF5 states with new phaser) but not forwards compatible (can't open new HDF5 states with old phaser).

[mlz-EM]

For future? When reconstruction runs on a CUDA-backed array module and drop_nans removes at least one raster pattern, mask is a device array while scan_meta['raster_rows'] and raster_cols are NumPy arrays. Indexing those arrays with ~mask attempts to convert the device mask to NumPy and fails (for example, PyTorch reports that CUDA tensors cannot be converted directly), so NaN filtering breaks on GPU raster scans. Convert the mask to a host NumPy boolean array for the metadata indexing while retaining the backend mask for scan and pattern arrays.