Fix two problems with previous commit.
First, I can't use relonami to compare symbols, because relonami
doesn't have different values for different symbols until pass 3. I
was wrongly pairing a hi16 or ha16 with a different symbol's lo16.
Fix by comparing item pointers. Hack mach/proto/as/comm7.c to pass
each item pointer to mach/proto/as/mach5.c
Second, I broke hi16[...] and ha16[...] for absolute values. Fix by
checking hr_head with DOTVAL and DOTTYP.
Having fixed the second problem, I can now switch li32 to use
eval_hl() and emit_hl(), without breaking programs that li32 an
absolute value.