Speed up integration tests with masternodes

[codablock]

See individual commits. Most speedup comes from reducing the number of blocks required to activate DIP3 and from copying the datadir of the first node to all other masternodes, which reduces initial sync time.

This PR currently also includes #2641 which I'll remove after merge.

[UdjinM6]

Looks good (let's see if travis thinks so too :)) but I wonder if we could just tweak regtest params globally (in chainparams.cpp) and reuse cache (setup_clean_chain == False) instead of reimplementing data copy etc.

[UdjinM6]

utACK

[UdjinM6]

Some stats from my local machine.

Before:

TEST                     | PASSED | DURATION

autoix-mempool.py        | True   | 557 s
dip3-deterministicmns.py | False  | 325 s
llmq-signing.py          | True   | 357 s
p2p-autoinstantsend.py   | False  | 426 s
p2p-instantsend.py       | False  | 468 s

ALL                      | False  | 2133 s (accumulated)

Runtime: 682 s

(False == block sync simply failed at some point)

After:

TEST                     | PASSED | DURATION

autoix-mempool.py        | True   | 155 s
dip3-deterministicmns.py | True   | 460 s
llmq-signing.py          | True   | 99 s
p2p-autoinstantsend.py   | True   | 179 s
p2p-instantsend.py       | True   | 128 s

ALL                      | True   | 1021 s (accumulated)

Runtime: 460 s

👍