Fix and test mangling

[maleadt]

I was finally fed up with some complex signatures not rendering nicely in the profiler, but it took a couple of hours to squash all the bugs. We can now successfully mangle the following:

julia> GPUCompiler.mangle_sig(Tuple{GPUArrays.var"#35#37", CUDA.CuKernelContext, CuDeviceMatrix{Float32, 1}, Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{2, CUDA.DeviceMemory}, Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}}, typeof(Base.literal_pow), Tuple{CUDA.CuRefValue{typeof(^)}, Base.Broadcast.Extruded{CuDeviceMatrix{Float32, 1}, Tuple{Bool, Bool}, Tuple{Int64, Int64}}, CUDA.CuRefValue{Val{2}}}}, Int64})
"_Z3_3515CuKernelContext13CuDeviceArrayI7Float32Li2ELi1EE11BroadcastedI12CuArrayStyleILi2E12DeviceMemoryE5TupleI5OneToI5Int64ESA_E11literal_powS7_I10CuRefValueI1_E8ExtrudedIS2_S7_I4BoolSH_ES7_IS9_S9_EESD_I3ValILi2EEEEES9_"

shell> c++filt _Z3_3515CuKernelContext13CuDeviceArrayI7Float32Li2ELi1EE11BroadcastedI12CuArrayStyleILi2E12DeviceMemoryE5TupleI5OneToI5Int64ESA_E11literal_powS7_I10CuRefValueI1_E8ExtrudedIS2_S7_I4BoolSH_ES7_IS9_S9_EESD_I3ValILi2EEEEES9_
_35(CuKernelContext, CuDeviceArray<Float32, 2, 1>, Broadcasted<CuArrayStyle<2, DeviceMemory>, Tuple<OneTo<Int64>, OneTo<Int64> >, literal_pow, Tuple<CuRefValue<_>, Extruded<CuDeviceArray<Float32, 2, 1>, Tuple<Bool, Bool>, Tuple<Int64, Int64> >, CuRefValue<Val<2> > > >, Int64)

Includes tests based on demumble, a c++filt-like tool.

[maleadt]

@pxl-th @jpsamaroo Is the AMDGPU.jl segfault here known?