Improve performance of pvUnsplit #4
Description
One of the current performance bottlenecks is pvUnsplit when called on large slices. The implementation is rather naive and uses the list monad to find all possible reductions of the child slice. There might be faster ways of computing all reductions wrt. both asymptotic and real runtime.
Ideas:
- precompute reduction options for each note?
- better way to detect incompatibilities early?
- explicit construction of all legal combinations (instead of generate and test)?
- save allocations by not using list monad?