Switch to Loopy Kernel Callables

This issue keeps track of the big refactoring around kernel callables.

Most important TODO's:

• Switch to latest version of loopy
• Replace C instructions with kernel callables
• Replace CallInstructions with kernel callables

Sumfact TODO's:

• Revisit kernel interface of sum factorization kernels
• Directly call sum factorization kernels
• SIMD vectorization as kernel fusion (issue in itself)