Fix StructureMapUtilities failures by updating the StructureMap runtime in R4, R4B, and R5.

[jurmous]

This addresses two related runtime problems:

• named groups were only resolved from direct imports, not through transitive imports
• element-model targets could fail when assigning a non-primitive datatype into a choice element such as Extension.value[x] during anonymous backbone supplementation

This issue was identified while investigating the downstream HAPI report hapifhir/hapi-fhir#7690.

Problem

StructureMapUtilities.resolveGroupReference(...) stopped at the first import layer. That meant a map could not invoke a group that existed only in an imported map of an imported map.

StructureMapUtilities.processTarget(...) also mishandled some element-model assignments when the produced value was a non-primitive datatype. In practice this broke cases such as assigning a Reference into Extension.value[x] while materializing an extension under an anonymously supplemented backbone element.

Changes

• add recursive group lookup across the full import graph, with cycle protection and duplicate-match detection preserved
• normalize converted non-primitive datatype values before calling setProperty(...) on element-model targets
• apply the fix consistently in:
  • org.hl7.fhir.r4
  • org.hl7.fhir.r4b
  • org.hl7.fhir.r5
• add regression tests for:
  • transitive imported group resolution
  • anonymous backbone supplementation with Extension.valueReference

Impact

After this change:

• dependent groups such as CopyHumanName resolve correctly through transitive imports
• assignments like src -> tgt.value = create('Reference') as t then CopyReference(src, t) materialize correctly in the element model for anonymous backbone supplementation scenarios

Testing

Added focused regression coverage for R4, R4B, and R5.

References

• hapifhir/org.hl7.fhir.core#2360
• hapifhir/hapi-fhir#7690