Improve compiler error recovery for null coalesce assignment operator

[DoctorKrolic]

Redo of #74218
Fixes #73894

[AlekseyTs]

Done with review pass (commit 1), tests were not looked at. It looks like the change affects scenarios that fall outside of the stated goal of this PR. I suggest making adjustments to affect only error scenarios with target typed new on the right. #Closed

[DoctorKrolic]

This PR is based on the following semantics of ??= operator:

• If left type is a reference type, the right-hand side should also be that reference type
• If left type is nullable value type, the right-hand side should be an underlying value type of that nullable type

This semantics is independent on what shape the right-hand side is. Therefore, limiting this recovery to only new() scenarios seems like an overfitting to me. If you look at tests, I heavily verified other target-typed expressions to ensure they get assigned a type since they are most likely to benifit from this change in IDE scenarios even if specific bugs where not discovered yet. But even for non-target-typed scenarios this recovery makes sense since now compiler can insert additional conversions when necessary, which are gonna be present in the success scenarios anyway. Look at ErrorRecovery_ReferenceType_ImplicitConversion and ErrorRecovery_NullableValueType_ImplicitConversion for such examples

[AlekseyTs]

If left type is nullable value type, the right-hand side should be an underlying value type of that nullable type

I do not think this is accurate. #Closed

[AlekseyTs]

Done with review pass (commit 2), tests were not looked at. Some of the previously made comments still look relevant. Also, an alternative way to suppress cascading diagnostics was suggested. #Closed

[AlekseyTs]

@DoctorKrolic It looks like PR validation is failing #Closed

[AlekseyTs]

LGTM (commit 10)

[AlekseyTs]

@dotnet/roslyn-compiler For a second review on a community PR

[AlekseyTs]

@dotnet/roslyn-compiler For a second review on a community PR

[AlekseyTs]

@DoctorKrolic Does this fix #73894 or there is remaining work there? If there is a remaining work, could you please provide details in the issue?

[DoctorKrolic]

It seems that scenario from the issue related to other compound assignment operators might not be valid. If so then this PR fully closes the issue. Asked for clarification from the issue's author

[DoctorKrolic]

Ok, confirmed, other scenarios are not valid. Modified PR description to close the issue when it is merged

[AlekseyTs]

@DoctorKrolic Thank you for the contribution.