Improve performance when exporting Instances

[mint-thompson]

Fixes #1426 and completes task CIMPL-1242.

Instance export remains one of SUSHI's most time-consuming operations. This PR includes several changes that aim to reduce the amount of time needed to export Instances.

• When exporting an Instance, one of the first steps is to find the StructureDefinition that SUSHI is exporting an instance of. This StructureDefinition will usually have its elements unfolded during Instance export in order to validate the paths of assignment rules. These unfolding operations can be time-consuming. In order to avoid repeatedly unfolding the same elements, StructureDefinitions are reused by the InstanceExporter. This offers significant performance gains in cases where a project defines many Instances of the same StructureDefinition, and those Instance definitions include many assignment rules to the same paths.
• Before any of an Instance's rules are applied, its implied properties from its StructureDefinition are set. But, Instances of core FHIR resources will never have their own implied properties. The only case where there may be properties to set is when the Instance contains assignment rules on extension paths where the extension's URL is given as the slice name. Therefore, when the StructureDefinition is a core FHIR resource, other paths do not need to be checked for implied properties.
• When validating an assignment rule on an Instance, a frequent operation is finding an element's parent or children. This can be accomplished by searching the StructureDefinition's list of elements based on the starting element id. However, this can become expensive when the list of elements is long. Elements now maintain references to their parents and children in order to minimize the amount of times the full element list is searched or filtered.

As examples of the benefits of these changes, here are a few selected repos from the regression set, with run times for SUSHI 3.8.0 and for this change set. These times are wall-clock times from my personal system, which are not perfectly accurate measurements of performance. However, I feel that the magnitude of the time reductions are significant.

Repo	Number of Instances	SUSHI 3.8.0 time (seconds)	PR time (seconds)
HL7/fhir-saner#master	204	164	114
HL7/v2-to-fhir#master	247	114	68
joofio/test-epi-composition#main	2714	2362	290

In the regression set for this PR, some changes appear. However, these are due to existing bugs that were revealed by the changes in this PR. Two issues have been added to the SUSHI backlog for these bugs: https://standardhealthrecord.atlassian.net/browse/CIMPL-1248 and https://standardhealthrecord.atlassian.net/browse/CIMPL-1249

[mint-thompson]

One other thing I noticed is that the ElementDefinition.assignFshCode tests take a lot longer to run on this branch than on master.

Okay, so the good news is, it's not just you: this is also happening for me. This makes me think it's worth investigating, so I'll... investigate.

[mint-thompson]

I've made some small updates to comments. There's also a new issue for yet another bug discovered when investigating the differences in output on this branch. Next is investigating assignFshCode.