clause grafting is slow

[gares]

Internally we use insertion sort, but even worse we use it on the whole prgoram, not just the predicate in question.
Note that there are use cases for grafting after another predicate (coq-elpi tc solver), but they are not essential.

[gares]

This is greatly improved in #265 since we insertion-sort only withing clauses of the same predicate and, depending on the indexing directives/modes, in the same bucket.