Tracking issue: Make sure universal resolution has great error messages

[konstin]

Explaining a resolution error for a failed universal resolution is more complex than for regular resolution: A error may now be constraint to a specific fork, and the derivations may only happen due to the markers and requires-python bound of this fork. We need to check, fix and add tests for conflicts in universal resolution.

[charliermarsh]

@konstin: "The error messages are as good as the regular error messages, with the exception of #5046 which we should at least understand"

@BurntSushi: "One weird thing: the dependency relationship between a==1 and a==1{marker} is exposed in the error messages themselves. If you know that, and you read the error message, it makes sense." (Let's create a separate issue for this.)

[BurntSushi]

"One weird thing: the dependency relationship between a==1 and a==1{marker} is exposed in the error messages themselves. If you know that, and you read the error message, it makes sense." (Let's create a separate issue for this.)

That's here: #5618