optimize test of withNull

[TysonMN]

In hedgehogqa/fsharp-hedgehog#154, @cmeeren suggested that the test of withNull was inefficient. Here are the runtimes of this test reported by the Test Explorer in Visual Studio.

• When running just this test: ~250 ms
• When running all tests: 7-17 ms

In fact, this test is among the fastest tests in the test suite.

This PR rewrites this test to do exactly what is desired by the test. It does not generate unnecessary values and it does not do any retests with shrunk values. It is easy to verify that this test passes for the right reason because removing the call to withNull causes the test to fail. Here are the runtimes reported by the Test Explorer in Visual Studio.

• When running just this test: 166-141 ms
• When running all tests: 6 ms

This rough analysis shows that this test that was already among the fastest got a little bit faster.

My rewrite of this test contains more code. It also contains more lines although some of them could be combined. Another disadvantage is that Hedgehog.Random might disappear one day soon. This will change the API, but I think the functionality will remain.

On the positive side, I think this implementation is more idiomatic.

[cmeeren]

Thanks. I'm wary of merging this since 1) there's no important performance benefit, 2) it's more complex (requires knowledge about more Hedgehog stuff that at least I don't know anything about), and 3) you mention that Random might disappear. So I suggest we close this without merging. What do you say?

[TysonMN]

What do you say?

That's fine. At least it is a specific proof of concept for an alternative implementation.