Rework blockchain import

[holiman]

This PR addresses some shortcomings in the blockchain downloader and block processing, specifically the behaviour around (invalid) long sidechains, as happened during the Ropsten constantinople HF.

[holiman]

This has now been running both full and fast sync on the benchmarking machines, and appear to be able to complete a full sync.
The graph below shows the fast-sync network IO after fast sync completion. Whereas ingress is almost the same, the ingress traffic is reduced heavily, almost to 50%. I believe that is because when we sync with a new peer, we now correclty determine the common ancestor at block N instead of block N-15, and do not have to fetch 15 full blocks unnecessarily.

The graphs also shows that block processing is faster in this PR -- but that's not really fair, because this PR handles some errors before 'starting the clock' -- and we only really measure the time execution of a block takes.

[karalabe]

@holiman Could you please rebase this on master? Thx

[holiman]

done

[karalabe]

An initial request set of comments. I still need to look through part of the core in detail, but there's quite a bit of refactoring I requested which modified the code heavily.

[holiman]

I think I have addressed all concerns now. However, after the latest changes, the code is relatively untested. Once you are satisfied with the general 'look' of the code, we should either add testcases that cover long (bad-state) sidechains, or run some external tests on that scenario.