Miner should not dial client on restart

[dirkmc]

Fixes #4991 #5211 #5224 #5268 #5038

• Error out if there are too many consecutive restarts with no data transferred limit consecutive restarts with no data transfer go-data-transfer#129
• Assess feasibility of writing an integration test / testground test
• Manual test
  It wasn't a small task to write an integration or testground test, so this should be tested manually:
  • Bring up a client and miner
  • Start transferring a large file from client to miner
  • While the file is transferring, restart the miner
  • Ensure that the file completes transferring and the deal moves to the sealing state

[magik6k]

(assuming that there's some good reason that this is a draft)

[dirkmc]

👍 It's a draft at the moment because I want to try to write some integration tests today to make sure the retry logic works

[dirkmc]

Yesterday we performed a manual test.
The data transfer will restart correctly if the client or the miner restarts once during the transfer.
If the client or miner restarts twice during the transfer, the second time the data rate monitor in go-data-transfer may miss the restart because of ipfs/go-graphsync#141

However I think this is a big improvement, and there is also improved logging so we can see exactly what the client and miner side of the data transfer are doing. So I favour merging this PR and then bubbling up improvements to graphsync as they are released.

For future reference, to see all the relevant data transfer activity, the following log subsystems should be set at debug level:

• markets
• dt-pushchanmon
• dt-impl
• data_transfer_network