[train][ci] Fix tune_hvd_keras import order

[justinvyu]

Why are these changes needed?

The test is failing due to the ordering of horovod and tensorboardX.SummaryWriter imports. This is a known issue: pytorch/pytorch#30651. This PR fixes it by moving the horovod import into the training loop -- to happen after the tensorboardX import on the driver.

The tests was passing before the xgboost_ray version bump, so there's probably some underlying package that got updated that's the root cause.

Reproduction of the bug

Horovod first, then tensorboard fails with a segfault:

>>> import horovod.keras as hvd
2023-09-01 10:13:20.236734: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2023-09-01 10:13:20.236757: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
>>> from tensorboardX import SummaryWriter
2023-09-01 10:13:29.194509: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2023-09-01 10:13:29.194539: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
Segmentation fault

Tensorboard first then horovod works:

>>> from tensorboardX import SummaryWriter
>>> import horovod.keras as hvd
2023-09-01 10:14:36.773923: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2023-09-01 10:14:36.773949: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.

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 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 :(