Use scippnexus to save nexus files

[YooSunYoung]

Currently we are using h5py in quite a few place.
We should probably use scippnexus for consistency.

essnmx/src/ess/nmx/nexus.py

Lines 361 to 362 in be53f9c

	def _export_static_metadata_as_nxlauetof(
	experiment_metadata: NMXExperimentMetadata,

essnmx/src/ess/nmx/nexus.py

Lines 398 to 399 in be53f9c

	def _export_detector_metadata_as_nxlauetof(
	*detector_metadatas: NMXDetectorMetadata,

Especially here we need compression for counts so that requires some update in scippnexus.

It can be useful for other instruments as well: #130 (review)

essnmx/src/ess/nmx/nexus.py

Lines 431 to 494 in be53f9c

	def _export_reduced_data_as_nxlauetof(
	dg: NMXReducedDataGroup,
	output_file: str | pathlib.Path | io.BytesIO,
	*,
	append_mode: bool = True,
	compress_counts: bool = True,
	) -> None:
	"""Export the reduced data to a NeXus file with the LAUE_TOF application definition.
	Even though this function only exports
	reduced data(detector counts and its coordinates),
	the input should contain all the necessary metadata
	for minimum sanity check.
	Parameters
	----------
	dg:
	Reduced data and metadata.
	output_file:
	Output file path.
	append_mode:
	If ``True``, the file is opened in append mode.
	If ``False``, the file is opened in None-append mode.
	> None-append mode is not supported for now.
	> Only append mode is supported for now.
	compress_counts:
	If ``True``, the detector counts are compressed using bitshuffle.
	It is because only the detector counts are expected to be large.
	"""
	if not append_mode:
	raise NotImplementedError("Only append mode is supported for now.")
	with h5py.File(output_file, "r+") as f:
	nx_detector: h5py.Group = f[f"entry/instrument/{dg['detector_name'].value}"]
	# Data - shape: [n_x_pixels, n_y_pixels, n_tof_bins]
	# The actual application definition defines it as integer,
	# but we keep the original data type for now
	num_x, num_y = dg["detector_shape"].value # Probably better way to do this
	if compress_counts:
	data_dset = _create_compressed_dataset(
	name="data",
	root_entry=nx_detector,
	var=sc.fold(
	dg['counts'].data, dim='id', sizes={'x': num_x, 'y': num_y}
	),
	chunks=(num_x, num_y, 1),
	dtype=np.uint,
	)
	else:
	data_dset = _create_dataset_from_var(
	name="data",
	root_entry=nx_detector,
	var=sc.fold(
	dg['counts'].data, dim='id', sizes={'x': num_x, 'y': num_y}
	),
	dtype=np.uint,
	)
	data_dset.attrs["signal"] = 1
	_create_dataset_from_var(
	name='time_of_flight',
	root_entry=nx_detector,
	var=sc.midpoints(dg['counts'].coords['t'], dim='t'),
	)

[SimonHeybrock]

Relevant section in the docs for making an application definition for this purpose: https://scipp.github.io/scippnexus/user-guide/application-definitions.html#Writing-files.

Note however that ScippNexus does not provide a lot of benefit/convenience over plain h5py when writing. It can therefore be ok to simply stick with h5py.

[jl-wynen]

I originally brought this up. my reason was that we may want to use the same compression scheme in other techniques, too.