Use block tracker alongside headers pool

[SebastianMarian]

• Fixed an edge case situation when self headers were removed from headers pool after a forced roll back action. This caused sometimes that computeSelfNotarizedHeaders from block tracker, which had used only headers pool for searching headers, to not found/return them. Furthmore, registered call handlers which should been notified about self notarized headers extracted from a received cross header were not called too, as these headers were not been found in headers pool.

Testing steps:

a) Prerequisites

b) Test steps
Start a normal testnet, with forks, restarts and txs

c) Expected results
Shards and meta nodes should recover faster after a forced roll back (action which cleans headers pool above the roll back nonce). We can grab then in logs after message: "roll back one block has been forced"
and check after how many rounds nodes have been proposed the next block (it should be around 1 or 2 rounds)

d) Actual results
Test process is in pending

[LucianMincu]

System tests passed.