Fix eltype conversion for StepRangeLen{BigFloat} (fixes #23300)

Ref #23348
(cherry picked from commit 1c2e689)

Author: timholy

base/twiceprecision.jl

@@ -422,7 +422,7 @@ function rat(x)
     y = x
     a = d = 1
     b = c = 0
-    m = maxintfloat(narrow(typeof(x)))
+    m = maxintfloat(narrow(typeof(x)), Int)
     while abs(y) <= m
         f = trunc(Int,y)
         y -= f
@@ -435,6 +435,7 @@ function rat(x)
     return a, b
 end
 
+narrow(::Type{T}) where {T<:AbstractFloat} = Float64
 narrow(::Type{Float64}) = Float32
 narrow(::Type{Float32}) = Float16
 narrow(::Type{Float16}) = Float16

test/ranges.jl

@@ -921,3 +921,10 @@ let linsp = linspace(1.0, 2.0, 10)
     @test Float32(linsp.ref) === convert(Float32, linsp.ref)
     @test Float32(linsp.ref) ≈ linsp.ref.hi + linsp.ref.lo
 end
+
+# Issue #23300
+x = -5:big(1.0):5
+@test map(Float64, x) === -5.0:1.0:5.0
+@test map(Float32, x) === -5.0f0:1.0f0:5.0f0
+@test map(Float16, x) === Float16(-5.0):Float16(1.0):Float16(5.0)
+@test map(BigFloat, x) === x