draft -- make parse_insert() more than 2x as fast

[gatoatigrado]

sorry this is benchmarked on some private data ... using a simple script to dump to JSON,

gato:~/sandbox/insert-parser-comparison> time python parse.py 10x | md5sum 846d416e653c08b3f1914d4f53f573e9 - python parse.py 10x 1.97s user 0.33s system 99% cpu 2.309 total md5sum 0.03s user 0.01s system 1% cpu 2.309 total

gato:~/sandbox/insert-parser-comparison> time python parse.py 10x | md5sum 846d416e653c08b3f1914d4f53f573e9 - python parse.py 10x 1.01s user 0.04s system 99% cpu 1.064 total md5sum 0.03s user 0.01s system 3% cpu 1.063 total

I'll try to productionize it at some point. I didn't test with complete=False and stuff.

[gatoatigrado]

and ... the _group() trick doesn't work on linux, just mac ... def. another reason not to push this yet! :)