Incorrect bounds computation in `index_from_s3.py`

[Kirill888]

1. MTL document lists extents as they apply to pixel centers, but datacube expects extents that cover whole raster edge to edge

datacube-dataset-config/scripts/index_from_s3_bucket.py

Lines 78 to 84 in 5d20cf6

	def get_geo_ref_points(info):
	return {
	'ul': {'x': info['CORNER_UL_PROJECTION_X_PRODUCT'], 'y': info['CORNER_UL_PROJECTION_Y_PRODUCT']},
	'ur': {'x': info['CORNER_UR_PROJECTION_X_PRODUCT'], 'y': info['CORNER_UR_PROJECTION_Y_PRODUCT']},
	'll': {'x': info['CORNER_LL_PROJECTION_X_PRODUCT'], 'y': info['CORNER_LL_PROJECTION_Y_PRODUCT']},
	'lr': {'x': info['CORNER_LR_PROJECTION_X_PRODUCT'], 'y': info['CORNER_LR_PROJECTION_Y_PRODUCT']},
	}

So recorded span is 15 meters (half a pixel) smaller than actual span.

2. The usual error of incorrectly computing bounding box in Lon/Lat domain that assumes that projecting bounding box to Lon/Lat is the same as computing bounding box after projecting the whole shape into Lon/Lat with extra points added and then computing the bounding box

datacube-dataset-config/scripts/index_from_s3_bucket.py

Lines 87 to 94 in 5d20cf6

	def get_coords(geo_ref_points, spatial_ref):
	t = osr.CoordinateTransformation(spatial_ref, spatial_ref.CloneGeogCS())
	def transform(p):
	lon, lat, z = t.TransformPoint(p['x'], p['y'])
	return {'lon': lon, 'lat': lat}
	return {key: transform(p) for key, p in geo_ref_points.items()}