[dbnode] Decode perf time ops

[rallen090]

What this PR does / why we need it:

Decode operation optimizations such as changing time.Time to int64 operations. The 4 rightmost pink blocks are time operations, each ~10ms, which have been removed.

Speed improvements look to be about +22%

Master
BenchmarkNextIteration/series_2000-12                210           5555359 ns/op
BenchmarkNextIteration/series_2000-12                194           5877416 ns/op
BenchmarkNextIteration/series_2000-12                216           5561167 ns/op
BenchmarkNextIteration/series_2000-12                214           5574842 ns/op

All changes
BenchmarkNextIteration/series_2000-12                273           4294894 ns/op
BenchmarkNextIteration/series_2000-12                267           4312589 ns/op
BenchmarkNextIteration/series_2000-12                277           4481239 ns/op
BenchmarkNextIteration/series_2000-12                261           4534195 ns/op

Special notes for your reviewer:

Does this PR introduce a user-facing and/or backwards incompatible change?:

Does this PR require updating code package or user-facing documentation?:

[arnikola]

Do you think it's worth checking that 0 <= u < len(s) or is that being a bit too defensive?

[rallen090]

Hmm yeah - I'm also not totally sure. We always new the schemes array via newTimeEncodingSchemes which ensures the array is size of unitCount, which means we know u will be valid in it. But since technically TimeEncodingSchemes could be created manually maybe it is safest to just check to be especially safe against an invalid index err.

[arnikola]

LGTM /w 2 small nits