MAINT: `stats.ContinuousDistribution`: adjustments per review

[mdhaber]

The first commit addresses the first round of comments:

• renames iv_policy to validation_policy per reviewer preference
• eliminates variable oo in favor of inf per reviewer preference
• reduces test time by changing the number of examples checked by Hypothesis (default was 100; now 20)

The second commit is the response to the concerns about the rng attribute.

Unfortunately, I think all the fuss was for nothing. I mis-remembered how things were working, and the current behavior is as we discussed it should be during the meeting on Wednesday at 11 a.m.: unless provided, a new Generator is instantiated every time it is needed. The implications for a ContinuousDistribution object are very straightforward:

• if no random state is provided, behavior of a ContinuousDistribution object is nondeterministic, and any copies are entirely independent.
• if a random state is provided, behavior of a ContinuousDistribution object is deterministic, and the behavior of any copies will be identical to that of the original.

Let me know if I should include that text in the docstring.

I expanded a test to check this behavior.

Note that I never allowed ints to be passed for rng because the argument is called rng, not seed. So if a user accesses the attribute rng, they either get a Generator if one was provided or None otherwise; they would never get an integer.

@tupui @steppi @ev-br @tirthasheshpatel @chrisb83 @dschmitz89

[mdhaber]

Thanks. Let's add that when SPEC7 merges and SciPy approves. I think it's better if we don't push changes to that code; showing that only the test has been strengthened.

[mdhaber]

The third commit changes uses of "array" -> "array_like" in documentation to address https://github.com/scipy/scipy/pull/20834#pullrequestreview-2175097036.