Endleaf rejiggle, .fsm syntax for endids, refactor endid get/set api, various related bugfixes

[katef]

I did quite a few things here, all focused around the endleaf callbacks,
and by extension the endid mechanism. I fixed a few small bugs, and got distracted trying to tidy up a bit.

Anyway here's what we get:

• .fsm syntax support for endids
• we now have tests for .fsm syntax (gasp)
• Add missing printing for endids for various formats (notably dot and .fsm)
• Various work on ways to simplify the endids API, especially:
  • Removed &nwritten for fsm_endid_get(), we don't need partial gets
  • Removed the tri-state enums for fsm_endid_get/set()
  • We don't need to distinguish a state absent endids from a present but empty set of endids
• Add missing calls to endleaf() callbacks for various formats
• Reintroduced fsm_exec handling in re(1), and reworked related argv[] handling. This is neccessary bugfixes only, in particular I want to address Behavior of -G length inconsistent, moreover returns error code. #478 and Unclear how to do multiple transformations in one invocation #477 separately
• Bugfixes around the api format codegen, that was quite out of date.

The .fsm syntax for endids looks like this:

; ./build/bin/re -r pcre -bpl fsm 'ab?c' 'abc*'
0  ->  1 "a"; # e.g. "a"
1  ->  2 "b"; # e.g. "ab"
1  ->  3 "c"; # e.g. "ac"
2  ->  4 "c"; # e.g. "abc"
4  ->  5 "c"; # e.g. "abcc"
5  ->  5 "c"; # e.g. "abcc"

start: 0;
end:   2 = [0], 3 = [0], 4 = [0, 1], 5 = [0];

and renders out like:

[silentbicycle]

Do you have a better suggestion for how to pass a handle to the caller besides void *opaque that isn't significantly more complicated in a multi-threaded context?