[repr types] eliminate noop costs (lowering impl)

[mgree]

The varchar_to_text cast is functionally a noop, but exists to satisfy the typechecker: varchar(8) and varchar and text are different SQL types.

But they are not different representation types---all of these will be Datum::String at runtime. So: in MIR, we can eliminate this cast.

This PR alters lowering (gated behind the feature flag enable_cast_elimination) to not bother generating varchar_to_text when lowering HIR to MIR.

Alternative implementation to #34348, which fails because the non-repr typechecker gets upset (and because I didn't rewrite the tests yet).

Both PRs will need to have only the repr typechecker running to avoid inconsistencies #34351.

Motivation

• This PR adds a known-desirable feature.
  design doc: MIR typechecking using representation types #27239

Checklist

• This PR has adequate test coverage / QA involvement has been duly considered. (trigger-ci for additional test/nightly runs)
• This PR has an associated up-to-date design doc, is a design doc (template), or is sufficiently small to not require a design.
• If this PR evolves an existing $T ⇔ Proto$T mapping (possibly in a backwards-incompatible way), then it is tagged with a T-proto label.
• If this PR will require changes to cloud orchestration or tests, there is a companion cloud PR to account for those changes that is tagged with the release-blocker label (example).
• If this PR includes major user-facing behavior changes, I have pinged the relevant PM to schedule a changelog post.

[ggevay]

My earlier attempt to do this (#27029) failed in part due to the adapter crate also calling MIR's typ in several places, to get RelationDescs. Has the repr types workstream progressed far enough in the meantime to have eliminated this issue?