Revert "use generics in random (#16283)"

[ringabout]

This reverts commit 71e2a9e.

[cooldome]

Please leave description on why. Useful for history records at least.

[timotheecour]

[EDIT]

the reason is #16283 introduced undesirable breaking changes as was discovered during post review of #16283 and #16288:

import std/random
proc main()=
  type Foo = 1..5
  for i in 0..<10:
    let b = rand(Foo.high)
    echo (b, $b.type)
main()

1.4.0:
(5, "int")
(0, "int")
...
current devel 1.5.1 2297b96:
(5, "Foo")
Error: unhandled exception: value out of range: 0 notin 1 .. 5 [RangeDefect]

proposal

this will make everything sane again, and makes 0 compromises on features:

the API prior to #16283 was sane, impossible to misuse, because the input range matched the output range:
proc rand*(max: int): int
proc rand*(max: float): float =

after #16283 it stopped being sane and has either unfixable edge cases (because input range != output range), or breaks existing valid code.

So I recommend:

• merging this which will revert back to the API from before use generics in random #16283
• then make rand consistent with other overloads #16288 will be rebased to just keep its improvements to trandom.nim (turning a weak "action:compiles" into real tests)

After that we can fix the remaining issues:

• fix the bugs with rand(HSlice) (rand(a: HSlice) fails with large ranges of int64 or uint64 values #16296)
• extend proc rand*[T: SomeInteger](t: typedesc[T]): T = to work with T: Ordinal instead of just T: SomeInteger

Note that rand(HSlce) allows doing everything you want (eg: rand(0..uint64.high) or rand(MyEnum.low .. MyEnum.high)), so it turns out we actually don't need to support rand(uint64.high)