Repair non-robust tests

[ablaom]

Some tests have had to be marked as broken, with the most likely culprit breaking changes to Distributions - for example Distributions 0.25.60 is known to change the behaviour of sampling for truncated distributions.

But the broken test and many others like it need reviewing as they are poorly designed. Look at for example this test from test/update.jl:

        @test_broken state.pbest_X ≈ [0.35710007620140427 0.37056803307520936 0.2723318907233864  2.7429797605672808 3.9372495283243105 -0.8067647083847199;
                               0.471258004384426   0.3400938657534834  0.18864812986209056 2.3976392099947    3.6569395920512977  0.420991611378423 ;
                               0.09661447846912713 0.19630556211862715 0.7070799594122457  2.5742724788985445 3.6354556967115146  0.6736019046580138]

There's no explanation as to where the numbers on RHS come from, our how to reproduce them. If they depend implementation details on Distributions, then this is not good, as Distributions seems likely to change from time-to-time and we have no way of knowing how to change these tests to address that.