Add support for PostgreSQL's trigram operators

[Niols]

Adds parsing and typing for PostgreSQL's pg_trgm operators:

• Similarity operators <%, %>, <<%, %>> → new TEXT_CMP_OP (comparison, Comp_text_cmp).
• Distance operators <->, <<->, <->>, <<<->, <->>> → new TEXT_DIST_OP (returns float).
• The similarity, word_similarity and strict_word_similarity functions (text × text → float).

Note: I deliberately left out the % trigram similarity operator. The difficulty is that % is overloaded across dialects: it's the integer modulo INT -> INT -> INT in both MySQL and PostgreSQL, but in PostgreSQL it also doubles as the trigram similarity operator TEXT -> TEXT -> BOOL. I don't think there is currently a good way to treat such a case, but we would need to have:

• Some form of union type for operators within the same dialect (eg. PostgreSQL's %s should have type (INT -> INT -> INT) U (TEXT -> TEXT -> BOOL)).
• Have different types for the same operators in different dialects, or even a way to make operators available only for one dialect.

I left it as a FIXME in the lexer, and I'm willing to work on this some time if you think this is worth pursuing.

[ygrek]

i was going to say i would expect some typing changes involving new ctor, but then i see there is nothing like this for comp_num_cmp either. Not for this PR but it means this is a missed opportunity to have better inference @jongleb ?

[ygrek]

as for % ambiguity i think we already have some ambiguity handling but it is case by case handling, also it will be the first one to depend on dialect i think yes