Improve bloom filter test

[huaxingao]

I saw bloom filter test Assert.assertFalse("Should not read: ...", shouldRead) failed with false positive.

This PR is to make the bloom filter test less flaky by taking consideration of fpp.

[szehon-ho]

While this makes sense, I wonder if its even worth to assert that number of false positives is less than some number? Each test asserting there's no false negatives may be good enough?

I assume that there is still a failure chance, and it may not really be suitable for unit test to be run each time? Curious what other parquet or other projects do. Also wondering what others think.

[huaxingao]

I think besides asserting no false negative, we probably still need to have a negative test for each of the data types?

[szehon-ho]

Yea it doenst seem great to have a chance of random failures in build tests but could go either way on it. Also curious with your buffer what is the estimated chance of failure ? (like is it astronomically small )

Wonder any thoughts from @kbendick @rdblue @RussellSpitzer

[rdblue]

Rather than adding code for false positives, wouldn't it be better to make the test use a consistent random seed that doesn't have a false positive? You'd need to replace UUID.randomUUID with a UUID built from random longs instead, but I think that would be better.

[huaxingao]

@rdblue Thanks for the suggestion! Done.

[rdblue]

Thanks, @huaxingao!

[huaxingao]

Thank you very much! @rdblue @szehon-ho