fix(cmSketch test) and add better cmSketch test

[FrankReh]

One Sketch test was intended to check that the seeds had caused each sketch row to be unique but the way the test was iterating, the failure wasn't going to be triggered.

With this fix, the test passes seeming to indicate the seeds are
working as intended on the row creation. But there is a problem.

A new Sketch test is added that goes to the trouble of sorting the counters of each row to ensure that each row isn't just essentially a copy of another row, where the only difference is which position the counters occupy.

And the new Sketch test is performed twice, once with high-entropy hashes, because they come from a PRNG, and once with low-entropy hashes, which in many cases will be normal, because they are all small numbers, good for indexing, but not good for getting a spread when simply applied to a bitwise XOR operation.

These new tests show a problem with the counter increment logic
within the cmSketch.Increment method which was most likely
introduced by commit f823dc4a.

A subsequent commit addresses the problems surfaced. But as the
discussion from issue #108 shows (discussion later moved to
https://discuss.dgraph.io/t/cmsketch-not-benefitting-from-four-rows/8712 ),
other ideas for incrementing the counters were considered
by previous authors as well.

Fix existing cmSketch test and add improved cmSketch test

(Marked as draft because this introduces a test that fails for now. I can commit a fix to the cmSketch increment method to get the new test to pass - if a maintainer agrees there is a problem to be fixed. See #108. I tried a few years ago.)

One Sketch test was intended to check that the seeds had caused each sketch row to be unique but the way the test was iterating, the failure wasn't going to be triggered.

With this fix, the test passes seeming to indicate the seeds are
working as intended on the row creation. But there is a problem with
the actual increment method. A new Sketch test is added that goes
further and sorts the counters of each row to ensure that each row
isn't just essentially a copy of another row, where the only
difference is which position the counters occupy.

And the new Sketch test is performed twice, once with high-entropy hashes, because they come from a PRNG, and once with low-entropy hashes, which in many cases will be normal, because they are all small numbers, good for indexing, but not good for getting a spread when simply applied with a bitwise XOR operation.

These new tests show a problem with the counter increment logic
within the cmSketch.Increment method which was most likely
introduced by commit f823dc4a.

A subsequent commit addresses the problems surfaced. But as the
discussion from issue #108 shows (discussion later moved to
https://discuss.dgraph.io/t/cmsketch-not-benefitting-from-four-rows/8712),
other ideas for incrementing the counters were considered by
previous authors of the original Java code as well.

Problem

Solution

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[github-actions]

This PR has been stale for 60 days and will be closed automatically in 7 days. Comment to keep it open.

[FrankReh]

It's too bad things like this aren't being fixed. I've seen the logic error copied as part of other language ports.

[github-actions]

This PR has been stale for 60 days and will be closed automatically in 7 days. Comment to keep it open.

[FrankReh]

Anybody?

[FrankReh]

I hope it is clear to anyone reviewing that this test is failing because of the logic bug I pointed out years ago. I had introduced a separate PR to fix that. This PR just highlights the problem more clearly for someone new.