[ci][tune][train] Update release test compute configs to not schedule work on head node

[justinvyu]

Why are these changes needed?

This PR updates the compute configs for benchmark release tests to not schedule workers onto the head node. This reflects the best practice not to schedule heavy work on the head node for cluster stability.

Test results for the flaky air_benchmark_tune_torch_mnist.aws:

train_times = [278.915740425, 269.3746311839999, 271.06450245400015]
train_mean = 273.1182913543334
tune_times = [291.6967464280001, 292.411954035, 290.2445501280001]
tune_mean = 291.4510835303334

291 / 273 = 1.07 << 1.35 (test pass threshold, previously the test was giving around ~1.33 due to the single trial on the head node being a straggler)

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 :(

[justinvyu]

See https://buildkite.com/ray-project/release/builds/24695