Warp sync state machine refactoring

[tomaka]

Close #67
Close #119
cc #864

[tomaka]

It's unclear to me how to manage malicious sources the correct way.
Presently we do it with a very naive banning system where one error results in a ban, and all bans are reset if all sources are banned.

[tomaka]

The difficulty in designing many aspects of the warp sync state machine comes from #1109

In this PR, I'm going to remove the concept of "successful warp sync" from the warp sync state machine, and instead modify the AllSync to treat the first successful warp sync as a trigger to switch to "all forks syncing". This way, #1109 consists only in modifying AllSync.

[tomaka]

It's unclear to me how to manage malicious sources the correct way.

I think the best approach is to first query only the source(s) with the highest finalized block, but if requests fail progressively broaden the search to all sources.

The high-level code must also be involved as it should kick out sources that don't answer properly.