MSB may be cleared in conversion from `BigFloat` on ARM

[kimikage]

julia> BigFloat(typemax(UInt64))|>N63f1 # x86-64 / i686 
1.8446744073709552e19N63f1 

julia> BigFloat(typemax(UInt64))|>N63f1 # Aarch64 / ARMv7
9.223372036854776e18N63f1

Originally posted by @kimikage in #200 (comment)

The following workaround is harmful to BigFloat.

FixedPointNumbers.jl/src/normed.jl

Lines 330 to 333 in 3f12729

 # a workaround for 32-bit ARMv7 (issue #134) 
 function _unsafe_trunc(::Type{T}, x::AbstractFloat) where {T} 
     unsafe_trunc(T, unsafe_trunc(typeof(signed(zero(T))), x)) 
 end