argmax for tensors with custom type index and AVX2 optimization (mex)

This MEX function provides the argmax functionality in Matlab for the purpose of avoiding the syntax of the max function from Matlab
[~,Y] = max(X,[],dim)
In addition it allows to return the indices in a user specified type (e.g. int32) and not just the default double.
Speed: when using -march=native in machines with AVX2 it allows interesting speedups in comparison to Matlab (except for double). Using AXV2 256bit registers it is possible to compute the maximum in parallel over elements of 2,4,16 or even 32 for types respectively double,float/int32,int16 and int32. The interesting part is the propagation of the indices because a AVX2 max is trivial. For using this feature it is necessary to pass -march=native to mex (e.g. modifying the XML configuration).

Added comparison of the results using the indices: result from Matlab and this could could differ in indices if the matrix contains duplicate values.

Usage:
Y = argmax(X, dim, int16(0)); % returns indices as int16

TODOs:
- min
- min and max in one pass
- check on dimension and specified type
- remake in C using Python for code generation