Fix for unstable behavior of HyperLogLog comparator

[turu]

• removed unsafe heuristics from hll compareTo as described in TopN query with sorting by hyperUnique metric returns badly ordered results #2447
• provided unit test for regression

Fixes #2447

[fjy]

@turu awesome catch. Do you mind trying the @ignore methods in HyperLogLogCollectorTest.java before and after the patch to verify performance?

[turu]

@fjy thanks :).

Here are the results (avg from 3 runs on my machine) for the ignored long running tests in HyperLogLogCollectorTest.java:
Before patch (at current tip of master):

1. testHighCardinalityRollingFold - 62s 583ms
2. testHighCardinalityRollingFold2 - 109s 594ms
   After patch:
3. testHighCardinalityRollingFold - 62s 882ms
4. testHighCardinalityRollingFold2 - 109s 513ms

[xvrl]

Thanks for the awesome bug report @turu we don't get many of those. 👍 on merging this from my end.

[gianm]

👍

[fjy]

@xvrl @gianm we should backport this

[turu]

Thanks for a swift reaction guys 👍