Overzealous resource reference validation

[lmsurpre]

Describe the bug
A "type constraint" on a Reference is one where you constrain the allowed types to which that Reference can refer. In our current implementation, these reference type constraints are not validated in our base model (#511).

However, for profiles that further constrain the allowed types of a reference, we're currently generating FHIRPath constraints that will cause validation errors unless the Reference.reference meets the following criteria:

• is present
• has a value that "looks like" its of the type resource; or has a value that we cannot infer a type from

I don't think this is right. While using a valued Reference.reference is certainly preferred, I don't think we want to prohibit references to identifiers or even references with just a display value.

To Reproduce
Steps to reproduce the behavior:

1. find or create a profile that constrains Reference
2. create a resource instance that has a Reference with an Reference.identifier but no Reference.reference

Expected behavior
Even though there is no Reference.reference, it should pass validation.