[optimize] MRE::typ(): can we avoid recalculating types?

[mgree]

Part of MirRelationExpr::typ() cleanup.

Is it possible to avoid recalculating types?

Motivation

• This PR adds a known-desirable feature.
  https://github.com/MaterializeInc/database-issues/issues/2488
  https://github.com/MaterializeInc/database-issues/issues/3779
  https://github.com/MaterializeInc/database-issues/issues/4959

Tips for reviewer

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.

[mgree]

Short answer: no. Recalculating the types gives us more precise keys.

assertion `left == right` failed
  left: RelationType { column_types: [ColumnType { scalar_type: Int32, nullable: false }], keys: [] }
 right: RelationType { column_types: [ColumnType { scalar_type: Int32, nullable: false }], keys: [[]] }

Trying one more run, this time just using the found_typ... what breaks?

[ggevay]

It's surprising that the key was wrong in the MirRelationExpr::Constant from which as_const got the type. We might want to investigate this (maybe independently from this PR).

[ggevay]

Oh, could you rebase on top of #30881?
I have a theory on what might have happened: take_safely used to not set the best possible key, because it ran the key derivation on the original thing, where the key derivation often has no chance to realize that it's an empty relation (because the caller of take_safely might have determined this in complex ways). If this doesn't change after the rebase, then there are other places where we don't accurately set the keys on constants.

[mgree]

In #30894.

[mgree]

Rebase is green... which means that we're still getting the less precise types on constants, right? 🤔