In `report_fullfillment_errors` push back `T: Sized`, `T: WellFormed`
and coercion errors to the end of the list. The pre-existing
deduplication logic eliminates redundant errors better that way, keeping
the resulting output with fewer errors than before, while also having
more detail.
Verify that `PartialEq` implementations do not break type inference
when comparing types across different allocators. This catches a
regression in current nightly introduced in 001b081cc1 (alloc: Allow
comparing `Box`s over different allocators")
`Box` is the only type that currently impelements this, but tests are
included for `Rc` and `Arc` to prevent future regresssions.
