Remove unreachable expand code from expr.c

[Paolo Bonzini]

> Hmm, while testing doesn't show any new FAILs, I guess this
> patch would break -O0 because CONJ_EXPR is only lowered when
> we're optimizing, ie. we never run pass_lower_complex at -O0.
> So forget about this one for now.

You're right.  As soon as Honza merges his patch to get the tree CFG 
(and thus bsi's) at -O0, I'll post one to enable a "pre-expand" pass 
that will run tree_complex at -O0.  Much code in optabs.c will also 
vanish then.

Paolo