fix(evmreader): avoid delaying work by skiping old blocks notifications

[renatomaia]

EVM Reader was processing each single new chain block notification, therefore when the processing was too slow (>2 seconds) the notifications accumulated and the EVM Reader could not keep up since worked one block at time. The fix is to skip old block notifications when a more recent is already available.

[mpolitzer]

UNTESTED

You can drain the channel without relying on block number retrieval:

for {
    select {
    case header = <-headers:
    default:
        break
    }
}

[renatomaia]

I tried that but the notification thread does not execute fast enough to keed the channel from blocking.
I added a size to the channel (buffer of notifications) to be able to mitigate this problem, but it should be a size large enough that the notification thread will be scheduled before our for discarding the notifications can finish.

Anyhow, @vfusco suggested I poll the block number and remove the use of single block notification.