New vectorization features

So far, only horizontal opportunities with identical input coefficient were considered. This merge request aims at more general vectorization opportunities. It involves:

• More involved SIMD transpose implementations
• Extension of the vectorization generator
• Generation of lower/upper half loads (like Vec8d(Vec4d, Vec4d))
• Introduce abstraction of kernel output (just like input)
• Handle horizontal_adds on SIMD halves
• Trim down variability of vectorization opportunities through heuristics
• Reimplement fromlist strategy on top of the new style

For the time being, I will restrict myself to lower/upper half of the input SIMD vector being different, as I can imagine this particular case to be very useful in vectorizing skeleton integrals. Allowing too many opportunities will screw the opportunity generator complexity-wise I guess, but we will see....