Fix test for "do branches have different typeclass_infos ...

... for the same constraint?"

This *does* complete the workaround for Mantis bug #585.

compiler/simplify_proc.m:
    Do not forget the typeclass_infos bound in branched control structures;
    keep them around for the code *following* the branched control structure.
    This is essential for the test to work for any branched control structure
    in which the current one is enclosed.

tests/invalid/tci_liveness_abort.err_exp:
    Expect diagnostics for both predicates where Mantis bug #585's
    conditions arise.