To be clear: time.Duration is a type, and is the only part of this discussion where “type safety” is a factor.
time.Millisecond is a constant time.Duration whose value represents the duration of a millisecond. time.Millisecond et al are not a unit as in chemistry class, so you’re not supposed to get a time.Millisecond^2 by multiplying them. Just like in any other typed language, T * T -> T, not T^2 (where T = time.Duration), since it’s a type and not a unit.
time.Millisecond et al instead make you explicitly annotate the scale of user-provided durations. When you write
d := time.Second*time.Duration(input)
you are assigning to d a value of type time.Duration equal to the duration of one second multiplied by the input. You are free to not use the constants defined by the time package and instead create time.Durations from the literal number of nanoseconds in the duration, if you so choose: d := time.Duration(input)*1000000000
But since the time package already defines convenient constants for standard durations, those will be used instead and be far more explicit and readable.This has nothing to do with dimensional analysis of units, of the sort you do in physical sciences. I don’t know of any general-purpose languages that include dimensional analysis in the language or their standard library’s datetime package, but I’d be curious if you know of one (that isn’t specifically geared towards the sciences) - seems like the sort of thing Ada might include? In any case, I don’t think a language focused on keeping a simple feature set like Go should be a trailblazer here. Durations are easy.