[cgraph] Avoid depending on torch CPU module for CPU-only actor

[hipudding]

Why are these changes needed?

In cases where the actor involved in the Compiled Graph exclusively uses the CPU, AcceleratorContext unnecessarily loads torch’s CPU module, resulting in an avoidable dependency.
This PR eliminates that dependency by avoiding the torch module loading for CPU-only actors.

Related issue number

Closes #53716

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 :(

[Copilot]

Pull Request Overview

This PR updates AcceleratorContext to skip importing the torch CPU backend when the device is CPU-only, removing an unnecessary dependency on torch.cpu.

• Conditional import of torch.<backend> only for non-CPU devices
• Eliminates unconditional torch.cpu import for CPU-only actors

Comments suppressed due to low confidence (3)

python/ray/experimental/channel/accelerator_context.py:40

• The comment should mention how the CPU case is handled (e.g., that no torch backend is loaded or that self._torch_mod will be None). This will clarify expected behavior for maintainers.

# Import the torch backend module (e.g., torch.cuda) if the device is not 'cpu'.

python/ray/experimental/channel/accelerator_context.py:1

• Add a unit test for the CPU-only path to verify that AcceleratorContext initializes correctly without importing a torch backend and does not raise errors when accessing self._torch_mod.

def __init__(self, torch_module_name: str, communicator_cls: Type[Communicator])

python/ray/experimental/channel/accelerator_context.py:41

• When torch_module_name is "cpu", self._torch_mod is never defined, which may lead to an AttributeError later. Consider adding an else block to define self._torch_mod (e.g., None or the base torch module) or refactor usage to handle a missing backend module.

if torch_module_name != "cpu":

[hipudding]

@ruisearch42 @kevin85421 Could you please review this bug fix? Thanks.

[ruisearch42]

Thanks. Let me trigger the tests.

[ruisearch42]

https://buildkite.com/ray-project/release/builds/45740#

[jjyao]

Release test passed: https://buildkite.com/ray-project/release/builds/46021#019784e0-6688-4732-b8dc-2d2f81f2b1c6