TypeAnalysis speed up Canonicalize

[wsmoses]

Again doing a assert equal to old implementation first, then will remove and merge.

[ZuseZ4]

improves compile time by 2x to 3.5x (impact becomes larger for bigger vectors).
Will check at some later point if it brings compile times down to something linear.
Updating a vector with 100 entries goes from 2s to 1s, for a vector with 500 entries from 10s to 4s.

Checked some slightly more realistic (larger) functions and unfortunately this mostly seems to help when adding more small functions, less when having fewer larger functions.
E.g. having 5 functions with each update a subset of 500 vec entries only improves from 180s to 120s.

For scale, real molecules have at least in the 10k entries, easily 100k or even more.