[Sema] Fix assert with incorrectly allowing optional injection into operator inout argument.

[gregomni]

When matching types on an operator 'inout' param and lvalue argument, be more restrictive with the inner match by changing the match kind. Allowing optional injection here is a no no.

Resolves SR-8221.

[gregomni]

@swift-ci Please smoke test and merge.

[xedin]

For changes like these you’d want to run source compatibility suite and ask somebody e.g. @rudkx to take a look before merging...

[gregomni]

@xedin Alrighty. Is there a way to tell swift-ci not to merge after all? A quick look about the docs doesn't reveal anything to me..

[gregomni]

@swift-ci Please test source compatibility

[xedin]

Only if you force push while it’s running, that’s going to abort the job.

[gregomni]

FWIW, there shouldn't be any functional change here, this is only keeping the constraint system from coming up with an incorrect solution that then hits an assert in CSApply when it can't create a valid coercion for that wrong solution.

[xedin]

Understand but since users get builds without asserts it’s possible that this is going to fail one of the projects by either not producing any solutions of a different one, we has multiple times that happened recently when @rudkx tried to land similar changes.

[rudkx]

This seems like the right fix. inouts parameters and their arguments have to be of the same type.