Need zero weight nodes in reduced tree for chain tips after latest messages

[vladzamfir]

The current implementation misses edge cases where there are zero weight blocks that are children of the fork choice (as implemented), but are not the latest message of any validator.

Should add the zero weight tips that are later than any latest message from any validator to the reduced tree.

[naterush]

Questions:

1. can we still bound the size of the tree to the number of validators with these new blocks? It seems there can be many zero-weight blocks: on the order of |V|^2, or something...
2. @paulhauner had an idea that would allow us to keep less of the zero-weight blocks. To summarize (if I heard it correctly [he rudely whispered it during a presentation]): if a zero weight block is added on top of another zero weight block, we can delete the bottom one.
3. Can we possibly store these zero-weight blocks in a separate structure?