[RLlib] Cast types for new configurations in PB2.

[simonsays1980]

Fixed a bug in generating a new config in 'PB2'. The type for floats were actually 'numpy.dtype.float64' and resulted so far in error in 'ray.rllib.utils.schedule.Scheduler.validate()' where this type is not allowed for the learning rate,

Why are these changes needed?

PB2 as one of Ray major hyperparameter tuning algorithms for ML and specifically RL was erroring out when tuning the learning rate. This was due to the fact that ray.rllib.utils.schedule.Scheduler.validate is allowing only specific types for the learning rate. Whenever PB2 created a new config it assigned the values received from ray.tune.schedulers.pb2._explore and that was type numpy.dtype.float64. So, basically it was not possible to tune the learning parameter in experiments.

This PR fixes this shortcoming and casts types from a new config explored via Bayesian Optimization to the ones from the old config. Like this configs will keep types over the course of an experiment and PB2does not error out anymore.

Related issue number

Closes #42180

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• 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 added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• 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 :(

[sven1977]

LGTM (added one minor change); Thanks for this fix @simonsays1980 !

[sven1977]

Looks great. Waiting for tests to pass, then merge.