[tune] Adjust searcher sample bounds to match Tune API

[krfricke]

Why are these changes needed?

We're currently passing searcher bounds directly to search algorithms - however, while we e.g. define randint's upper bound to be exclusive, this is not the case for other search algorithms (e.g. Optuna).

This PR adds a more thorough testing framework to check the sampling behavior of all our searchers. It makes sure that the sampled values from any searcher is within the bounds defined by the Tune search space API. It also checks that (at least for discrete values) the search space is sampled extensively (i.e. that the highest and lowest discrete values are sampled, too, so that we didn't cut off too much of the search space).

Related issue number

Closes #15593

Checks

• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[richardliaw]

Looks fine but did not carefully review.

Can we add this note somewhere in the docstrings for the tune search space primitives? Perhaps as a changelog?

https://www.sphinx-doc.org/en/master/usage/restructuredtext/directives.html#directive-versionchanged