Revert #129

[vlovgr]

It seems the assumption made in #129 is incorrect: poll(timeout) with timeout > 0 can both result in assignment and new records being returned for the new assignment. Therefore, we generally need to store records before fetch requests are received. This pull request effectively reverts #129.

[codecov]

Codecov Report

Merging #194 into master will decrease coverage by 0.23%.
The diff coverage is 76.27%.

@@            Coverage Diff             @@
##           master     #194      +/-   ##
==========================================
- Coverage   91.86%   91.62%   -0.24%     
==========================================
  Files          57       57              
  Lines        1376     1397      +21     
  Branches       54       54              
==========================================
+ Hits         1264     1280      +16     
- Misses        112      117       +5

Impacted Files	Coverage Δ
...e/src/main/scala/fs2/kafka/internal/LogEntry.scala	26.47% <0%> (+1.47%)	⬆️
.../scala/fs2/kafka/internal/KafkaConsumerActor.scala	84.07% <78.94%> (-0.48%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0b78ee1...8a5153e. Read the comment docs.