Make FH allocation-free #91
Description
We should use in place solves and preallocated workspaces instead of allocating at each iteration, i think this is the reason why FH is so slow.
Ultimately, all the NLPModels API functions for the FH problem should be allocation free.