consistently elide/don't elide types in diagnostics

right now, the decision of whether to fully elide the type as _ is inconsistent - it never happens for functions, fn pointers, tuples, refs where the pointee types are equal, or adts, but it does happen for primitives and ADTs behind a raw pointer. that is probably not intended? might be worth moving the t1 == t2 comparison up above the big match for consistency. https://github.com/rust-lang/rust/blob/eb53721a34d1d910b900aa753e0a00dc72ef41ac/compiler/rustc_infer/src/infer/error_reporting/mod.rs#L1510-L1512

here is an example of it being inconsistent: https://github.com/rust-lang/rust/blob/cf2dff2b1e3fa55fa5415d524200070d0d7aacfe/tests/ui/issues/issue-17905-2.stderr#L23-L26
isize is elided but str is not.

i originally changed this to almost always elide the type, but @estebank is worried that will be confusing:

he suggested instead eliding if the type has either type params or projection types ("looks complicated", basically).

Originally posted by @jyn514 in https://github.com/rust-lang/rust/issues/118730#issuecomment-1846520771

@rustbot label +A-diagnostics +E-medium

Start in compiler/rustc_infer/src/infer/error_reporting/mod.rs around lines 1510-1512, then compare the diagnostic output in tests/ui/issues/issue-17905-2.stderr. Resolve the intended type-elision rule from the issue's competing proposals, implement consistent behavior, and update or add diagnostic coverage showing the expected output for the listed type forms.