Use data_type and nodata from raster extension if present

[TomAugspurger]

Currently the default dtype is float64 and the default fill_value is np.nan.

The raster extension defines fields for specifying the dtype (data_type) and fill value (no_data). It might be a good idea to change the default to using values from raster:band, falling back to float64 / np.nan if the raster extension isn't present.

[gjoseph92]

Thanks for opening the issue to track this; there's some more info in #63 (comment) too. Definitely excited about the raster extension! @TomAugspurger think you'd have any time to work on this?

[TomAugspurger]

Thanks for the link to #63.

We're going to standardize our use of the raster extension in our STAC API over the couple months. Once that work is started I'll be happy to work on this. I just wanted to make sure there was agreement with the API change (to use information from raster:bands) before starting on that.

[gjoseph92]

Awesome, happy to hear raster:bands is going to get some traction!

Note that in doing this, we'll probably want to add some special-handling logic for raster:bands, similar to how the subfields of eo:bands get pulled out into top-level coordinates (I could also see making this more generic to any dict-ish fields in flattened_metadata_by_asset):

stackstac/stackstac/prepare.py

Lines 485 to 518 in 3a58906

	eo_by_asset = []
	for meta in flattened_metadata_by_asset:
	# NOTE: we look for `eo:bands` in each Asset's metadata, not as an Item-level list.
	# This only became available in STAC 1.0.0-beta.1, so we'll fail on older collections.
	# See https://github.com/radiantearth/stac-spec/tree/master/extensions/eo#item-fields
	eo = meta.pop("eo:bands", {})
	if isinstance(eo, list):
	eo = eo[0] if len(eo) == 1 else {}
	# ^ `eo:bands` should be a list when present, but >1 item means it's probably a multi-band asset,
	# which we can't currently handle, so we ignore it. we don't error here, because
	# as long as you don't actually _use_ that asset, everything will be fine. we could
	# warn, but that would probably just get annoying.
	eo_by_asset.append(eo)
	try:
	meta["polarization"] = meta.pop("sar:polarizations")
	except KeyError:
	pass
	coords.update(
	accumulate_metadata.metadata_to_coords(
	flattened_metadata_by_asset,
	"band",
	skip_fields={"href"},
	# skip_fields={"href", "title", "description", "type", "roles"},
	)
	)
	if any(eo_by_asset):
	coords.update(
	accumulate_metadata.metadata_to_coords(
	eo_by_asset,
	"band",
	fields=["common_name", "center_wavelength", "full_width_half_max"],
	)
	)