eth: don't enforce minimum broadcast, fix broadcast test

[karalabe]

Currently we always propagate a block/transaction to at least 4 peers, even if the square root is less (e.g. if I have 5 peers, I will propagate to 4, not just to 2). As far as I know this was added when there was no announcement mechanism at all (i.e. eth/60).

I think this is misguided:

• If a node is so unhealthy that it barely has a few connections, there's no point in pushing even more load onto it.
  • Every node will surely broadcast to at least 1 more peer (sqrt(whatever_gt_1) >= 1) and the announces will cover any gaps for super degenerate topologies.
• If a node is healthy, it will always happen (when broadcast saturates the network) that most of its peers already sent it a block/tx, so PeersWithoutBlock and PeersWithoutTx will return a small subset of peers. With the enforced 4 peer propagation, the node will send the block/tx forward to 4 further peers, even if 46 out of 50 peers already know of it, which is pointless.
  • Announces should handle degenerate connections, everything else can still be saturated with sqrt(whatever) (since sqrt(whatever_gt_1) >= 1, so we will always send the block in its entirety to at least one peer).

The PR also reduces the test time of the BroadcastTest from 18 seconds to 1.