Prefetch descriptor functions at graph-handling time, not resource-handling time

[jacobtylerwalls]

#10439 applied in two more places the trick from #10312 to skip queries for descriptor functions on the assumption that resources are generally grouped by graph:

arches/arches/app/utils/data_management/resources/formats/archesfile.py

Lines 393 to 395 in 1af6ea8

	# Reuse the queryset for FunctionXGraph rows if the graph is the same.
	if last_resource and (last_resource.graph_id == this_resource.graph_id):
	this_resource.descriptor_function = last_resource.descriptor_function

But we could get more savings by removing this "trick" and just prefetching the descriptor functions before we start iterating the resources.

That would go something like this:

• When querying for resources, use a Prefetch() object to do the filtering for the graph's descriptor function and save it to descriptor_function on each Resource.
• Then remove the "trick" linked above in both files it appears

Then, with the hood popped on this (looking at callers of save_descriptors()), there are a few more wins to be had:

• index_resources_by_type() refetches graph objects because the earlier call in index_resources() only gets a values array
• Tile.delete() refetches the resource instance
• Tile.delete() saves the resource descriptor once per node
• import_business_data_without_mapping() fetches graphs one-by-one

[jacobtylerwalls]

The last few bullet-points tacked on here don't provide enough value to work on.

index_resources_by_type() refetches graph objects because the earlier call in index_resources() only gets a values array
Tile.delete() refetches the resource instance
Tile.delete() saves the resource descriptor once per node
import_business_data_without_mapping() fetches graphs one-by-one

[1] -- not very many queries
[2] -- refetching is necessary, because we need the Proxy model
[3] -- already fixed in #10481
[4] -- too hard to refactor