[Breaking] Change default rem rounding mode to RoundToZero

The previous default was `RoundNearest`, which was inconsistent with the
behavior for Base float types, as noted in issue 112. The move to
`RoundToZero` is potentially subtly breaking but worthwhile for
behavioral consistency with Base.

Fixes issue 112

Author: ararslan

src/Quadmath.jl

@@ -346,9 +346,9 @@ end
     Float128(@quad_ccall(@autoql(atan2)(x::Cfloat128, y::Cfloat128)::Cfloat128))
 
 Base.Integer(x::Float128) = Int(x)
-@assume_effects :foldable Base.rem(x::Float128, y::Float128, ::RoundingMode{:Nearest}=RoundNearest) =
+@assume_effects :foldable Base.rem(x::Float128, y::Float128, ::RoundingMode{:Nearest}) =
     Float128(@quad_ccall(@autoql(remainder)(x::Cfloat128, y::Cfloat128)::Cfloat128))
-@assume_effects :foldable Base.rem(x::Float128, y::Float128, ::RoundingMode{:ToZero}) =
+@assume_effects :foldable Base.rem(x::Float128, y::Float128, ::RoundingMode{:ToZero}=RoundToZero) =
     Float128(@quad_ccall(@autoql(fmod)(x::Cfloat128, y::Cfloat128)::Cfloat128))
 
 if VERSION < v"1.14.0-DEV.2070"

test/rounding.jl

@@ -310,6 +310,10 @@ end
 end
 
 # These tests were copied from Julia. License is MIT: https://julialang.org/license.
+@testset "rounding consistency with Base" begin
+    @test rem(prevfloat(Float128(1)), Float128(1)) == prevfloat(Float128(1))
+    @test mod(prevfloat(Float128(1)), Float128(1)) == prevfloat(Float128(1))
+end
 @testset "rem(::Float128, ::Float128, ::RoundingMode)" begin
     T = Float128
     @test rem(T(1), T(2), RoundToZero)  == 1