check overflow in hex2num (fix #22031)

[rfourquet]

This also creates hex2num(::Type, str) to get a type-stable function, and fix (what I think to be) a bug, where num2hex(-1) == "-0000000000000001" instead of "ffffffffffffffff", i.e. num2hex is supposed to give the bit-pattern in hex form (similar to bits).

Edit: this is independant of the question on whether hex2num should be moved to parse...

[rfourquet]

I need here a function, say f, which takes a bits type and returns the corresponding Unsigned type with the same bit-width, so that e.g. f(Int) = UInt and f(Float16) = UInt16, to be used together with reinterpret. I first thougth of unsigned, but I feel unsigned(T::Type) should return the return-type of unsigned(x::T), i.e. unsigned(Float16(1)) == UInt(1) so unsigned(Float16) == UInt. I now chose reinterpret(Unsigned, T) == f(T) (e.g. reinterpret(Unsigned, Float16) == UInt16)) which does what is needed, but I'm not conviced... Any ideas for a better name for f?

[tkelman]

excess blank line here

I think "a hexadecimal" is more common than "an hexadecimal"

[fredrikekre]

[`hex2num`](@ref) maybe.

[tkelman]

should this be a separate docstring for the separate signature?

[rfourquet]

I don't know, but will do.

[tkelman]

the body of this for loop is indented one level too deep

[rfourquet]

Thanks @tkelman and @fredrikekre, comments addressed.

[fredrikekre]

Given #22088, is this still relevant?

[JeffBezanson]

It looks like this would still be useful if the hex2num-specific parts were removed.

[JeffBezanson]

Replaced by #22203