Fix racy batching on the dispatcher

[cyli]

TIL that go timers are racy from #2669. I noticed the dispatcher also calls Reset without ensuring that it's stopped or expired first, so added a similar fix. (See golang/go#11513)

dispatcher/heartbeat.go also has a Reset without calling Stop first, but this seems ok? If an extra heartbeat gets triggered, that seems inexpensive and fine?

[codecov]

Codecov Report

Merging #2676 into master will increase coverage by 0.05%.
The diff coverage is 0%.

@@            Coverage Diff             @@
##           master    #2676      +/-   ##
==========================================
+ Coverage   62.06%   62.12%   +0.05%     
==========================================
  Files         134      134              
  Lines       21740    21744       +4     
==========================================
+ Hits        13494    13508      +14     
+ Misses       6794     6779      -15     
- Partials     1452     1457       +5

[dani-docker]

@cyli
Good catch! this is the second instance #2669 where go timer messes up our logic.
We probably need to review every usage of timer in swarmkit

[dperny]

LGTM.

The code as-is should not be a major disruption. The worst case, as far as I can tell, is that we'll call Reset too late and end up immediately launching into another batch processing routine.

[thaJeztah]

ping @anshulpundir PTAL