Inconsistent int-float promotion in `hypot`

[danielwe]

hypot(::Int64, ::Float32) promotes the arguments and result to Float64, contrary to the behavior of promote, which gives Float32 in this case.

julia> hypot(1, 3.0f0)
3.1622776601683795

julia> promote(1, 3.0f0)
(1.0f0, 3.0f0)

The culprits are the following method definitions:

julia/base/math.jl

Lines 798 to 799 in 71f68b4

	hypot(x::Number, y::Number) = _hypot(promote(float(x), y)...)
	hypot(x::Number, y::Number, xs::Number...) = _hypot(promote(float(x), y, xs...))

Seems like this could be rewritten as follows for consistency with promote:

hypot(x::Number, y::Number) = _hypot(float.(promote(x, y))...)
hypot(x::Number, y::Number, xs::Number...) = _hypot(float.(promote(x, y, xs...)) 

[mikmoore]

Another consequence is permutation-asymmetry in the return type

julia> hypot(1, 1f0)
1.4142135623730951 # Float64

julia> hypot(1f0, 1)
1.4142135f0 # Float32