PATCH: use hashtab for pseudo op table

[DJ Delorie]

> I'd feel better about the interface, but I'd still be concerned
> about the constant factors.  And I want support for
> not-NUL-terminated strings with length passed in separately, too.

Perhaps we could hard-type the hash tables when we create them, and
have the hashtab internals "just know" about a few popular kinds of
hashes?  Then, when we create those types of hash tables, we can
optimize them better.

The default type would use callbacks, but specific types for
NUL-terminated strings, length/memory pairs, and whatnot could be
added too.  We just have to be careful not to break the existing API
when we change things.