Avoid should-fail in two ui tests

should-fail is only meant for testing the compiletest framework itself.
It checks that the test runner itself panicked.

tests/ui/borrowck/two-phase-reservation-sharing-interference.rs

@@ -3,7 +3,7 @@
 // The nll_beyond revision is disabled due to missing support from two-phase beyond autorefs
 //@ unused-revision-names: nll_beyond
 //@[nll_beyond]compile-flags: -Z two-phase-beyond-autoref
-//@[nll_beyond]should-fail
+//@[nll_beyond]check-fail
 
 // This is a corner case that the current implementation is (probably)
 // treating more conservatively than is necessary. But it also does

tests/ui/type-inference/box_has_sigdrop.rs

@@ -1,4 +1,4 @@
-//@ should-fail
+//@ known-bug: unknown
 //@ compile-flags: -Wrust-2021-incompatible-closure-captures
 // Inference, canonicalization, and significant drops should work nicely together.
 // Related issue: #86868