Speed discrepencies in dojo.query

[dante]

Running some slickspeed tests I noticed what could be a potential regression in dojo.query selector speeds. It seems to choke on all variations of p:nth-child(), pushing test results well into the red. The speeds are balanced until the test reaches those selectors. Attached are totals from a number of browsers testing the original and a modified version of slickspeed.

[dante]

my results, pseudo-csv format

[alex]

(In [16870]) fixes a typo in getNodeIndex which caused node indexes to be lost, dropping our speed on nth-child queries. Also significantly improves peformance on ":not" pseudos by specifying ignores in the test generator. Fixes #8654. !strict

[alex]

(In [16871]) even more speedups on the DOM branch. Was doing some dumb stuff for ID and tag-only queries which this checkin fixes. Also, attempt to preserve zip-killing across groups. Tests show us to be ~30% faster than sizzle on IE 6. QSA branch still faster nearly everywhere, but by variable amounts depending on the engine (15-50%). Refs #8654. !strict