Add Materialize Iterator

[barakmich]

Introducing the materialize iterator -- if it seems reasonable to just materialize a set instead of checking paths that might lead back, the And iterator can now add this layer as an optimization. It bails if used on sets that are too large, and is still ripe for cost/benefit tweaks, but in general will make deep queries a lot faster.

To wit, the benchmarks are now almost all short:

$ benchcmp master.txt aftermaterialize.txt
benchmark                                   old ns/op        new ns/op       delta
BenchmarkNamePredicate                      904393           898317          -0.67%
BenchmarkLargeSetsNoIntersection            150923975        54921214        -63.61%
BenchmarkVeryLargeSetsSmallIntersection     174235833178     463987963       -99.73%
BenchmarkHelplessContainsChecker            18569994238      15828325039     -14.76%
BenchmarkNetAndSpeed                        32027611         16516696        -48.43%
BenchmarkKeanuAndNet                        13430475         12509101        -6.86%
BenchmarkKeanuAndSpeed                      15260371         14882936        -2.47%
BenchmarkKeanuOther                         1133568511       74794130        -93.40%
BenchmarkKeanuBullockOther                  124273081626     84608986        -99.93%

/cc @pbnjay @kortschak

[kortschak]

LGTM with comment above. Nice work.

[pbnjay]

if it.hasRun, you can copy it.values and it.containsMap here. Since these aren't changed after they are populated you can even save the allocations and just copy the refs too.

[pbnjay]

LGTM!