Fix random polNodes sticking around

[kcalvinalvin]

This PR is related with #287

EDIT:

Second commit adds pollard size to the stats, making it easier to keep track of the pollard size.

[adiabat]

try:

if h!=0 {
    leftRoot.niece, n.niece = n.niece, leftRoot.niece          // swap
}

and for n.prune() it can take an h==0 bool as an argument, which if true gives this different functionality:

if n.niece[0].deadEnd() && n.niece[1].deadEnd() {
	n.niece[0], n.niece[1] = nil, nil
}

[dergoegge]

The last part of my comment here might explain why there are random nodes sticking around.
I think the simplest solution to remembering proofs is to use a dedicated polnode that leaves point to if they are supposed to be remembered.

[dergoegge]

I feel like creating a special rememberNode would be the easier patch with less side effects.
rememberNode could either be a global or a field on Pollard.

[dergoegge]

it would probably be better to set remember=false when swapping the leaves

[kcalvinalvin]

I did try to get that to work but TestCache() kept failing and I bet there's a bunch of edge cases to handle. I'm leaning more towards just getting this in and then working on setting remember to false during swaps since that's an optimization.

[dergoegge]

maybe creating a polLeaf type would be good because the remember bool is stored on all nodes now but really only the leaves need it.
Both polNode and polLeaf would need to inherit from the same interface.

[kcalvinalvin]

Hmm yeah. I also hate the fact that niece isn't nieces for roots and leaves. Maybe we can solve all these problems at once.

[adiabat]

OK - this fixes the bug but at the cost of increasing the memory usage around 17% (extra 8 bytes for the bool on all nodes).
We need to fix that space usage, but will merge this for now because it does fix the caching.
We might fix this by going to pointing to a (global) "memorable" dummy node to point to. Or some other larger fix to pollard reprsentation.