Warp syncing shouldn't be a one time process

[tomaka]

Imagine this situation: the light client starts, connects to a node, and warp syncs to block 500. It then switches to "all forks syncing".
Later, it connects to a node that advertises block 100000.

In that situation, the light client should try to warp sync to block 100000 from that other node while keeping in mind that this other node could be malicious.

Basically, the AllSync state machine should continue running the warp sync state machine even after it's done.