Implement cleanup of run files upon model session and run deletion

[Creylay]

Summary

Fixes a bug where Run.run_path on disk can be either a file (scikit-learn/torch models) or a directory (Hugging Face models). Since Run.id gets reused by SQLite once rows are deleted, a stale artifact left behind by one run could collide with a new run that lands on the same id, crashing training with IsADirectoryError/PermissionError. This same file-vs-directory assumption also broke DELETE /run/{id}: deleting a run whose run_path was a directory always failed with a 500, since it used os.remove (files only) instead of a helper that handles both. Also fixes a related cleanup gap where deleting a model session never removed its runs' files from disk, letting stale artifacts orphan in the first place.

---

Type of Change

Check all that apply like this [x]:

• Backend change
• Frontend change
• CI / Workflow change
• Build / Packaging change
• Bug fix
• Documentation

---

Changes (by file)

• DashAI/back/job/model_job.py: clear any pre-existing file/directory at run_path before saving a trained model (root-cause fix).
• DashAI/back/api/api_v1/endpoints/runs.py: delete_run now uses remove_p, so runs whose run_path is a directory (Hugging Face models) can actuallybe deleted.
• DashAI/back/api/api_v1/endpoints/model_sessions.py: delete_model_session now cleans up the run_path/plot files of cascaded runs on disk instead of leaving them orphaned.
• DashAI/back/api/utils.py: relocated the shared remove_path helper here el_job.py` can use it.
• tests/back/api/test_jobs.py, tests/back/api/test_runs_api.py: added regression tests for the three scenarios above.

---

Testing

• Manually reproduced the original bug (train a Hugging Face model → delete its model session → train new runs until an id is reused) and confirmed it no longer fails, and that the affected run can be deleted normally.
• Manually deleted a run whose run_path was a directory (Hugging Face model). Deletion now succeeds instead of returning a 500.