refactor(side-dag): general improvements

[glevco]

Motivation

Implement some of the requested changes in previous PRs.

Acceptance Criteria

• Remove poa.in_turn_signer_index() and add poa.get_signer_index_distance().
• Update poa.calculate_weight() to decrease the block weight inversely proportional to the index distance.
• Changes in PoaBlockProducer:
  • Merge both looping calls into one.
  • Add delayed block production cancellation.
  • Update logs.
  • Change block delay function to include a constant offset for out of turn blocks, and increase the interval between each signer.
• Update tests accordingly.

Checklist

• If you are requesting a merge into master, confirm this code is production-ready and can be included in future releases as soon as it gets merged

[codecov]

Codecov Report

Attention: Patch coverage is 83.67347% with 8 lines in your changes missing coverage. Please review.

Project coverage is 85.00%. Comparing base (7829afb) to head (c605607).

Files	Patch %	Lines
hathor/consensus/poa/poa_block_producer.py	80.48%	5 Missing and 3 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1080      +/-   ##
==========================================
- Coverage   85.08%   85.00%   -0.08%     
==========================================
  Files         314      314              
  Lines       23937    23957      +20     
  Branches     3616     3621       +5     
==========================================
- Hits        20367    20365       -2     
- Misses       2867     2879      +12     
- Partials      703      713      +10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[glevco]

I had to make some changes because I noticed the behavior was not working as expected, in relation to cancellation of block production. Before these changes,

• Since delayed calls were not cancelled when the delayed block was in turn, that meant we could have more than one active delayed call, and only one of them would have a pointer (in self._delayed_call).
• In fact, the block used to check the turn was incorrect, as it was using the delayed call argument which is the previous_block, not the new one being generated.
• We were not cancelling block production when a new best block arrived from another peer.

All these issues have been addressed in a separate commit, 37ce560.

The base branch was changed.

[msbrogli]

I feel that the full node would not recover from getting out of sync after it was already synced. From recovery, I mean that this full node would stop producing blocks.

[glevco]

The full node recovers as long as it receives a block from another peer. I added an assertion in an existing test to cover this, in 0d4fea6

[msbrogli]

I still don't understand why we need this LoopingCall. Anyway, I'm ok to get this PR merged and get this changed later because it has no impact on mainnet and testnet.

[glevco]

Using either a LoopingCall or a recursive callLater, separate from the main callLater logic (triggered reactively via pubsub) were the only good ways I found to implement this while covering the following requirements:

• A single full node should be able to produce blocks by itself.
• If block production throws during the "start producing" phase, production should be tried again. However, if it fails after a call via pubsub, it should not be retried.
• We have to consider that the call to can_start_mining() may throw an exception.

I created an issue to review this later: #1097