Fixed bug involving labels in GIMPLE_COND statements.

Fixed bug involving labels in GIMPLE_COND statements.

Turns out that once we're at the GIMPLE level, GIMPLE_CONDs should no
longer have labels.  The optimization routine that gets rid of
"forwarder" basic blocks is aware of this, so it does not update the
labes in the GIMPLE_COND when it redirects an edge.

The routine that cleans up unnecessary labels, however, sees these
labels, which may have been destroyed by the forwarder optimization,
and can crash as a result.