Float32/16 raised to an integer's typemin throws a DomainError

[mbarbar]

It shouldn't throw because there is generally no issue raising a Float to a negative power.

julia> x = Float32(1.1)
1.1f0

julia> y = typemin(Int64)
-9223372036854775808

julia> x ^ y
ERROR: DomainError with -9223372036854775808:
Cannot raise an integer x to a negative power -9223372036854775808.
Convert input to float.
Stacktrace:
 [1] throw_domerr_powbysq(::Float64, p::Int64)
   @ Base ./intfuncs.jl:262
 [2] power_by_squaring(x_::Float64, p::Int64)
   @ Base ./intfuncs.jl:286
 [3] ^(x::Float32, n::Int64)
   @ Base.Math ./math.jl:1280
 [4] top-level scope
   @ REPL[66]:1

julia> x ^ (y+1)
0.0f0

At some point there's a -exponent but -typemin(Int) == typemin(Int).

I've also opened #57463 but this goes through a different path (and anyway produce an incorrect result rather than an exception).

[NHDaly]

(this reproduces on master)