Snark works are not rebroadcast periodically

[georgeee]

Test case

1. Launch a network, send some transactions
2. Shutdown one of BPs for some time
3. Start the BP back
4. Wait for 15 minutes
5. Shutdown other BPs, leaving the previously downed BP alive
6. Wait for another hour

Expected result

Block production continues as usual, transactions are included

Actual result

Some transactions are not taken due to lack of snark work

Alternative testing

With Nathan's fix of snark pool it's sufficient to check that snark pools for all nodes are of equal size after step 4.