[syncing] Trigger block import by hash instead of number

[dmitry-markin]

#1686 fixes an issue with failed imports of forks #493 when import is erroneously triggered by block number comparison instead of parent hash comparison (see #493 (comment)). This leads to the syncing state machine believing it can extend the canonical chain, while in reality blocks belong to a fork and cannot be imported (and we need a few more blocks downloaded backwards in order to reach the forking point). This wouldn't be happening if import was triggered by parent hash comparison instead of merely block number comparison.

The fix #1686 is fine short-term, but in the long term the logic must be updated to trigger import by parent hash comparison instead of block number comparison. This issue is about implementing this logic and obsoleting #1686.