…t#52185)

<!-- Thank you for your contribution! Please review
https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before
opening a pull request. -->

<!-- Please add a reviewer to the assignee section when you create a PR.
If you don't have the access to it, we will shortly find a reviewer and
assign them to your PR. -->

## Why are these changes needed?

<!-- Please give a short summary of the change and the problem this
solves. -->

We observed the following error in deploying Ray LLM V0:

```

2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | return run_method(self, method, args, kwargs)
-- | -- | -- | -- | --
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/utils.py", line 2255, in run_method
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | return func(*args, **kwargs)
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/worker/worker_base.py", line 604, in init_device
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | self.worker.init_device()  # type: ignore
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/v1/worker/gpu_worker.py", line 120, in init_device
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | self.model_runner: GPUModelRunner = GPUModelRunner(
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/v1/worker/gpu_model_runner.py", line 110, in __init__
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | self.attn_backend = get_attn_backend(
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/attention/selector.py", line 95, in get_attn_backend
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | return _cached_get_attn_backend(Show Details
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/attention/selector.py", line 148, in _cached_get_attn_backend
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | attention_cls = current_platform.get_attn_backend_cls(
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | File "/home/ray/anaconda3/lib/python3.11/site-packages/vllm/platforms/cuda.py", line 270, in get_attn_backend_cls
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | and kv_cache_dtype.startswith("fp8"))
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | ^^^^^^^^^^^^^^^^^^^^^^^^^
U | 2025-04-09, 10:50:12.511 | worker |   | ip-10-0-97-109 | AttributeError: 'torch.dtype' object has no attribute 'startswith'

```

This is because vLLM has inconsistent logic when handling
`kv_cache_dtype`. For V0, it is a string and for V1, it is a
torch.dtype. And when the V0 and V1 configs are messed up, this issue
manifests.

Another recent change in vLLM is that it auto tries V1 and use V1
configs if possible. Therefore, when we are not using an environment
VLLM_USE_V1, Ray LLM defaults to V0 and vLLM may use V1. This PR fixes
the inconsistency by always explicitly passing the environment variable.

## Related issue number

<!-- For example: "Closes ray-project#1234" -->

## 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 :(

---------

Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
Signed-off-by: Steve Han <stevehan2001@gmail.com>