[Diagnostics] Remove FailureDiagnosis::diagnoseParameterErrors from CSDiag

[LucianoPAlmeida]

This removes some obsolete code involving argument-to-param from repairFailures.
Removing FailureDiagnosis::diagnoseParameterErrors from CSDiag and fixing the remaining edge cases that were still being diagnosed by that. All of them related to ambiguity and generic types.
e.g.

func f<T>(_ i: T, _ j: T) {}
let _ = f(Int(3), Float(2.5)) 

cc @xedin @hborla

[xedin]

@LucianoPAlmeida I'll try to take a look tonight, thank you!

[xedin]

Overall I think this is a way to go but since @hborla is working on changed to diagnoseAmbiguityWithFixes she should take a look as well.

[xedin]

I believe that could be generalized to all fixes. That is exactly something @hborla is about to open a PR for, it would cover not only argument mismatches but any ambiguity which involves the same fix kind at a shared location :)

[LucianoPAlmeida]

Do you mean the same callee location? because in those cases all the fixes are the same, but each one for an argument. e.g. In the case of

func f<T>(_: T, _: T) {}
f(Int(1), Float(1))

There is a fix located at arg #1 in the first solution which was attempted (Int, Int)->Void and another at #0

[hborla]

Yeah, the two fixes here have different locators so it's not quite the same case as N common fixes across N solutions

[hborla]

Actually, maybe the locator doesn't matter. With my changes here, I've implemented a special method on CSFix for diagnosing for ambiguity when the fix is "common" across all solutions, but I consider a fix "common" only when the locator is the same. Instead of assuming that the locator must be the same, we can let the particular fix kind decide what it means to be "common"

[LucianoPAlmeida]

Humm ... so in for these case for AllowTupleTypeMismatch and AllowArgumentMismatch fixes will be special cases where we could ignore locator for common to be able to disambiguate.

[LucianoPAlmeida]

Overall I think this is a way to go but since @hborla is working on changed to diagnoseAmbiguityWithFixes she should take a look as well.

Oh right, Thank you for taking a look @xedin :)

[hborla]

Thank you for porting this @LucianoPAlmeida ! We are so close to removing CSDiag completely, and we really appreciate you continuing to help out!

As Pavel mentioned, I'm rewriting diagnoseAmbiguityWithFixes here to involve fewer special cases, and the strategy has changed to be based on the differences between solutions. It's currently looking at differences in overload choices, but we can cover a case like

func f<T>(_ i: T, _ j: T) {}
let _ = f(Int(3), Float(2.5)) 

by looking at the type variable binding for generic parameter T across all solutions and report could not infer generic parameter 'T' if there is a difference. However, I think we can take your change as-is and I'll take care of covering this case with my new changes.

[xedin]

@swift-ci please smoke test