Suppress suggest try wrap when found is unresolved infer

[xizheyin]

Fixes #145634

In this example, &Some((1, 2)) is an &_, and we cannot know the exact type, so I think suppressing this suggestion is the most appropriate choice here. This is similar to / inspired by #145361.

I expanded TypeFlags to indicate whether a type contains a type with is_ty_var true.

The test with regression(#3680, #5358, #12552) are all ICE, and some are even from many years ago. Therefore, this change in PR should not have a significant impact on them?

Two commits shows the diff of the newly added test.

r? compiler

[rustbot]

This PR modifies tests/ui/issues/. If this PR is adding new tests to tests/ui/issues/,
please refrain from doing so, and instead add it to more descriptive subdirectories.

[dianne]

If you're checking for inference variables on the error path, would it make sense to do so in a way that doesn't require tracking them for all types on the happy path too?

I think the "found" type may also be very likely to have inference variables when it doesn't match the expected type, on account of how pattern type-checking works. The type of the constructor pattern is checked against the type of the scrutinee, without knowledge of what the constructor's type's generic arguments are; I think those are all left for type inference to handle. As such, I think something more targeted may be needed if the suggestion is to support types with generic arguments.

[fee1-dead]

it seems like if the scrutinee isn't a reference while the pattern clearly is, we could try suggesting removing the & from the pattern

[xizheyin]

@rustbot author