Nanosecond specialization for iso8601.Parse

[chriso]

This adds additional specializations to iso8601.Parse (#82):

name                 old time/op  new time/op   delta
Parse-4              16.9ns ± 1%   18.7ns ± 1%   +10.76%  (p=0.002 n=6+6)
ParseMilliseconds-4  18.2ns ± 0%   21.3ns ± 0%   +16.93%  (p=0.002 n=6+6)
ParseMicroseconds-4   191ns ± 0%     23ns ± 1%   -87.99%  (p=0.004 n=5+6)
ParseNanoseconds-4    197ns ± 0%     25ns ± 0%   -87.47%  (p=0.002 n=6+6)

[achille-roussel]

ParseMicroseconds-4   191ns ± 0%   192ns ± 0%     ~     (p=0.026 n=5+6)

Should we support variable precision of the sub-second value? It seems we could easily miss opportunities to use the optimized code paths if trailing zeroes are trimmed for example.

[chriso]

Yeah I realized after submitting this that the stdlib (and probably elsewhere) trims those trailing zeroes. I thought when formatting a time with RFC3339Nano that you'd get back exactly 9 digits after the decimal point. It should be easy enough to generalize this to handle any digits after the decimal point.

[chriso]

@achille-roussel see 470c959

[achille-roussel]

Looking great 👍