[RLlib] quick fix for learning rate schedule for APPO algorithm.

[gjoliver]

Why are these changes needed?

LearningRateSchedule must be initialized after base.init() call.
Longer term, base classes should rely as little as possible on class member variables.
E.g., all these base modules could have simply taken config as input when necessary.

Related issue number

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 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 :(

[avnishn]

will need to check whether the appo vtrace failing test is related or is flaking

[gjoliver]

will need to check whether the appo vtrace failing test is related or is flaking

yeah, good call, I am gonna do it now.

[gjoliver]

will need to check whether the appo vtrace failing test is related or is flaking

yeah, good call, I am gonna do it now.

doesn't seem related. the test also doesn't use lr schedule at all.