Define isnan for AbstractTime

[ianfiske]

Fix https://github.com/JuliaLang/Statistics.jl/issues/53

[mschauer]

Should adapt the doc string and widen the definition of the scope (perhaps “a number l” instead of “a floating point number”) introducing a fallback isnan(::Number) = false

[mschauer]

If widening isnan to number-likes is not welcomed, a local solution would be

x -> x isa AbstractFloat ? isnan(x) : false

but that would break code for x which aren't AbstractFloats but could be NaN-likes, for example NaN with units.

[nalimilan]

At #35989 I also felt that the absence of an isnan(::Number) fallback was a problem to write generic code.

But I don't think adding isnan(::TimePeriod) is the good fix for https://github.com/JuliaLang/Statistics.jl/issues/53, as we will keep getting an error for other types which don't define it. And if we do that manually for all types, we'd better define isnan(::Any) = false -- which would be inconsistent with iszero.

Can't we just check applicable or method_exists before calling isnan in quantile?

[ianfiske]

Good point @nalimilan -- I've opened https://github.com/JuliaLang/Statistics.jl/pull/54 instead. Closing this.