testGithub2245 in testPickers.cpp occasionally fails

[ptosco]

if you run the test 1000 times, e.g. by doing

( FOR /L %i IN (1,1,1000) DO ( ctest -I 159,159 -V ) ) > debug.log 2>&1

you will likely incur in a couple of exceptions.
Exceptions always occur in testGithub2245() on either line 45 or 52:

rdkit/Code/SimDivPickers/testPickers.cpp

Lines 35 to 64 in ff5266f

	void testGithub2245() {
	BOOST_LOG(rdErrorLog) << "-------------------------------------" << std::endl;
	BOOST_LOG(rdErrorLog) << "Testing github issue 2245: MinMax Diversity picker "
	"seeding shows deterministic / non-random behaviour."
	<< std::endl;
	{
	RDPickers::MaxMinPicker pkr;
	int poolSz = 1000;
	auto picks1 = pkr.lazyPick(dist_on_line, poolSz, 10, RDKit::INT_VECT(), -1);
	auto picks2 = pkr.lazyPick(dist_on_line, poolSz, 10, RDKit::INT_VECT(), -1);
	TEST_ASSERT(picks1 != picks2);
	}
	{ // make sure the default is also random
	RDPickers::MaxMinPicker pkr;
	int poolSz = 1000;
	auto picks1 = pkr.lazyPick(dist_on_line, poolSz, 10);
	auto picks2 = pkr.lazyPick(dist_on_line, poolSz, 10);
	TEST_ASSERT(picks1 != picks2);
	}
	{ // and we're still reproducible when we want to be
	RDPickers::MaxMinPicker pkr;
	int poolSz = 1000;
	auto picks1 =
	pkr.lazyPick(dist_on_line, poolSz, 10, RDKit::INT_VECT(), 0xf00d);
	auto picks2 =
	pkr.lazyPick(dist_on_line, poolSz, 10, RDKit::INT_VECT(), 0xf00d);
	TEST_ASSERT(picks1 == picks2);
	}
	BOOST_LOG(rdErrorLog) << "Done" << std::endl;
	}

Initially I thought it was due to poor quality random seeding (see http://www.pcg-random.org/posts/cpps-random_device.html), but replacing

    if (seed >= 0) {
      generator.seed(static_cast<rng_type::result_type>(seed));
    } else {
      generator.seed(std::random_device()());
    }

with

    if (seed >= 0) {
      generator.seed(static_cast<rng_type::result_type>(seed));
    } else {
      generator.seed(randutils::auto_seed_128{}.base());
    }

as recommended here: http://www.pcg-random.org/posts/simple-portable-cpp-seed-entropy.html didn't remove the occasional failures over 1000 runs.
So I realized that the occasional failures are not due to poor seeding but rather to the small sample, which makes getting two identical picks once in a while not so uncommon.
So I believe the best solution to the problem is to make the test more robust against occasional identical picks.
PR will follow soon.