Revert "Disable unreliable sync_large_checkpoints_testnet"

[teor2345]

Now that we've done #1222, we can re-enable these tests.

Reverts #1789

Closes #1791.

[teor2345]

This PR failed in an unrelated test, #1878 should fix it.

I'm re-running the tests to make sure that it's a rare failure.

[teor2345]

Second test passed, let's run again just to be sure.

[teor2345]

Yep, the latest test run also looks fine, I'm also going to run it locally just to be sure.

[teor2345]

I ran the zebrad acceptance tests --including-ignored with #1877 and #1878 in a loop on my local machine, until one of the tests failed.

For the first few test runs, I got:

3 successes before the first failure due to a bad set of initial peers from the DNS seeders (only 2 good peers)
28 successes before the first failure due to a bad set of initial peers from the DNS seeders (only 1 good peer)
4 successes before the first failure due to a bad set of initial peers from the DNS seeders (only 3 good peers)
25 successes before the first failure due to a bad set of initial peers from the DNS seeders (only 1 good peer)

That's about a 7% failure rate - I'm not sure if that's acceptable. But my local machine has a much higher latency and lower network speed than our CI, so CI could fail a lot less often.

If the rate ever gets too high, we can double the number of zcashd or Zebra peers that we have deployed on testnet. (Our network security fixes and DNS seeder redeployment should also increase the number of good peers in the set of seeder peers, see #1753 and the recent network security tickets.)