Re-implement compile-time tensor pack by calling libxsmm's IDENTITY function

[rengolin]

Since #565 we have the ability to use libxsmm calls in the compiler.

We're working on lowering tensor.pack into tpp.copy calls in a loop (#290) but the compile-time implementation (#336) is still running through loops and is really slow and not guaranteed to do the same thing as time passes.

We need to re-implement the compile-time pass to do the same as the code generated.

@KavithaTipturMadhu