Add assignability rules for when the target is a conditional type

[weswigham]

Inspired by looking into #27014
Fixes #26933
Steals from #27589

This adds assignability rules for relating types when the target is a conditional type. We only check for conditional type target assignability when:

• The target is not distributive or,
• The distributive conditional type's check type cannot be instantiated to never
  Given these circumstances, a type is assignable to the conditional in question if
• it is assignable to both branches of the conditional, or
• if the conditional type looks to be always true (yet has infer types, preventing simplification) and it is assignable to the intersection of the instantiated constraint of the true branch and the uninstantiated constraint (this allows for "independent" infer types to not interfere with true branch assignability, whilst still preserving the assignment-blocking part of the existential-like behavior of the infer types).

[jack-williams]

When you write:

it is assignable to the intersection of the instantiated constraint of the true branch and the uninstantiated constraint.

By instantiated are you referring to the instantiation of the infer parameters? Would it be possible to write a small example that shows the difference between the proposed behaviour, and the behaviour if you only consider the uninstantiated true constraint (as opposed to the intersection)?

I have a hard time translating the rules into concrete examples, so it would be greatly appreciated!

[dgreensp]

Does this address the unsoundness in #26945?

[sandersn]

This experiment is pretty old, so I'm going to close it to reduce the number of open PRs.