Feat/more multi miner fixes jude

[jcnelson]

This PR fixes several more bugs and performance regressions in the networking stack and relayer. These bugs were discovered in a six-node topology with three miners and three stacker-signer followers, where each node was paired to its own bitcoin node and each stacker-signer follower was paired to a single follower (see schematic below):

            `bitcoind 0` <-----> `bitcoind 1` <-----> `bitcoind 2` <-----> `bitcoind 3` <-----> `bitcoind 4` <-----> `bitcoind 5`
                 |                    |                    |                    |                    |                    |
                 |                    |                    |                    |                    |                    |
                 V                    V                    V                    V                    V                    V
             `stacks 0`  <----->  `stacks 1`  <----->  `stacks 2`  <----->  `stacks 3`  <----->  `stacks 4`  <----->  `stacks 5`
             `(miner) `           `(miner) `           `(miner) `           `(follower)`         `(follower)`         `(follower)`
                                                                                ^                    ^                    ^
                                                                                |                    |                    |
                                                                                |                    |                    |
                                                                            `signer 0`           `signer 1`           `signer 2`

Fixes included the following:

• Fixes the wanted tenure and available tenure calculations in the Nakamoto downloader at a reward cycle boundary. Before, the available tenure list could be empty on a reward cycle boundary, which could lead to a downloader stall.
• Standardizes the first-block-height calculation of a reward cycle in the Nakamoto downloader, leading to the deletion of about 100 lines of code.
• Fixes the confirmed/unconfirmed download state transition logic by removing a bunch of code.
• Fixes the Nakamoto tenure downloader to return the tenure-start block of the next tenure as well as the tenure's blocks, so that when the relayer stores them and the chains coordinator processes them, the tenure will be treated as "finished" by the chainstate DB (which the downloader requires in order to mark the tenure as "fully obtained").
• The relayer will now push StackerDB chunks in addition to the StackerDB state machine. It will forward chunks that it was able to successfully store, indicating that the chunk was novel and it's likely that the node's neighbors need it.
• The Nakamoto inventory state machine will perform a configurable burst of inventory syncs with its neighbors right after a sortition is detected. This helps the downloader quickly detect when a neighbor has the next tenure-start block, so if the node doesn't get the tenure-start block pushed to it, it will likely find it shortly.
• Fixes Nakamoto block push by removing the late-stage check to see if the block had already been stored and aborting if so. This was a bug because the caller will have already stored the block before the relayer has a chance to consider it. Instead, the relayer will assume the caller is only trying to relay the block because the caller determined that the block was novel, as had been the case all along.
• Fixes the Nakamoto main loop to honor burnchain.poll_time_secs.

Things I still need to do:

• [x] Test the wanted and available tenure calculation at a reward cycle boundary in a unit test
• [x] Test StackerDB sync in a unit test whereby A synchronizes blocks with B, and B pushes blocks to C (and C doesn't run the StackerDB synchronization)
• [x] Consolidate some debug output

[ASuciuX]

Adding this here as link reference for the action for when it will be completed and the output will be displayed.

[obycode]

@jcnelson you need a rustfmt.

[jcnelson]

cc @obycode @jferrant ping :pray:

[kantai]

Pushed bfdb6e4 -- this should fix the unit test failures. The matches!() macro invocations were backwards -- if you match on the enum variant first, it treats the variable afterwards as a binding. Cargo would warn against this, but unused variable warnings are disabled in a lot of the testing codebase especially.

[obycode]

doh! looks like we still need a formatting fix.

[blockstack-devops]

This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.