[RLlib] Fix "Cannot convert a symbolic Tensor (default_policy/strided_slice_3:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported"

[sven1977]

This PR fixes the tensorflow error message occurring when using a newer numpy version with RLlib and tf:

The error message is:

NotImplementedError: Cannot convert a symbolic Tensor (default_policy/strided_slice_3:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

Why are these changes needed?

Related issue number

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 :(