Refactor handling submits batch to avoid stuck submits

[poszu]

Previously round::batch was created on the first submit to the batch and deleted after the batch was flushed. However, there existed a corner case when the batch flush was scheduled but the submit was rejected (for example, a challenge already present in DB) leading to a situation when the batch was never flushed again.

An example of this happening on the devnet-405:

Some submits time out because the flush never happens.

This PR mitigates it by:

• scheduling the flush when the first challenge is already put in the batch,
• simplifying batch handling by keeping its instance and using batch.Reset() instead of nil-ing it and recreating.

[codecov]

Codecov Report

Merging #403 (b2596e0) into develop (38d99cd) will not change coverage.
The diff coverage is 100.0%.

@@           Coverage Diff           @@
##           develop    #403   +/-   ##
=======================================
  Coverage     74.9%   74.9%           
=======================================
  Files           23      23           
  Lines         1906    1906           
=======================================
  Hits          1429    1429           
  Misses         359     359           
  Partials       118     118           

Files Changed	Coverage Δ
registration/round.go	94.3% <100.0%> (ø)