Add ClusterLocal as cluster scheduler without remote connection

[reginabuehler]

Description and Context:
What and Why?

This PR introduces a local cluster scheduler (developed by @sbrandstaeter) that mirrors the existing Cluster scheduler’s PBS/Slurm + Dask jobqueue workflow, but runs in a local access setting (no remote SSH connection).
A few remaining pieces are still pending and listed below.

What's missing:

•  cluster integration test
• eliminate code redundancy to schedulers/cluster.py and utils/start_dask_cluster.py

This pull request introduces a new scheduler hierarchy to support the new local cluster scheduler. A new common parent class now contains the shared cluster logic, while the remote and local schedulers are implemented as subclasses.

Related Issues and Pull Requests

• Closes
• Related to

Interested Parties

@dharinib98: This branch would allow us to run QUEENS simulations locally on the cluster.

Note: More information on the merge request procedure in QUEENS can be found in the Submit a pull request section in the CONTRIBUTING.md file.

[bennoschoenstein]

Hi, I've been using this PR and wanted to share some feedback and two small additions I made.

Addition 1: job_extra_directives parameter

I needed to add a SLURM --constraint to restrict jobs to specific node types (only certain nodes have VTK support). Since ClusterLocal had no way to pass extra SBATCH directives beyond the computed --ntasks, I added a job_extra_directives parameter:

```python
scheduler = ClusterLocal(
...
job_extra_directives=["--constraint=skylake&thin"],
)
```

In _start_cluster_and_connect_client, the existing line:
```python
"job_extra_directives": [job_extra_directives],
```
becomes:
```python
"job_extra_directives": [job_extra_directives, *self.job_extra_directives],
```

Addition 2: slurm_logs/ subdirectory for SLURM logs

With many jobs the .err/.out files pile up directly in the experiment directory. I changed log_directory to write into a slurm_logs/ subdirectory:

```python
"log_directory": str(self.experiment_dir / "slurm_logs"),
```

Would you consider including either or both of these in the PR?

[reginabuehler]

@bennoschoenstein Thanks for sharing your experience with using this branch and your feedback.
Both additions are very valid points and should ease usability. As far as I see, both concern the current cluster scheduler in cluster.py as well. Since they are not directly related to this feature of a cluster scheduler without a remote connection, I can also see them being introduced in a separate pull request.

On the current status of this PR: It still needs quite some refactoring to be able to be merged due to code redundancy. This refactoring then also (most likely) makes it easier to include these additions in a general manner.
Therefore, I suggest to keep your suggestions in mind while refactoring and possibly open separate issues/PR for these suggestions afterwards.

[m-frey]

@reginabuehler Thanks for this very usefull contribution, works for me without any issues.