fix: drop literal quotes from scontrol Reason/Comment argv (fixes #2954)

[SIDDARTHAREDDY8]

Problem

In helm/slurm-cluster/slurm_scripts/check_runner.py, drain_node, comment_node and uncomment_node build the scontrol argv with literal quotes: Reason="{reason}" / Comment="{comment}".

Because subprocess.run gets a list (no shell), nothing strips those quotes, so Slurm stores them as part of the value. On the OK path, comment.startswith(reason_base) then never matches, so on_ok: uncomment/undrain never fire — a node keeps its [node_problem] comment after the link heals. Reproduced from the scontrol show node --json output in #2954.

Solution

Drop the literal quotes: Reason={reason}, Comment={comment}, Comment= (3-line change, verified by diff against main).

Testing

• Repro harness mocking subprocess.run (captures argv, simulates slurmctld storing elements verbatim): bug reproduced — stored values carry literal ", and uncomment stores '""' instead of clearing.
• Unit suite (test_check_runner_fix.py, stdlib unittest): 8/8 pass — 5 positive (quote-free storage, uncomment clears, full comment→heal→uncomment and drain→heal→undrain round-trips through the real run_check FAIL/OK paths with the startswith(reason_base) guard) and 3 negative (old code never clears on the OK path).
• No real Slurm or subprocesses were used; only subprocess.run was mocked.

Release Notes

N/A (bug fix, no user-facing change beyond correct node recovery).

[SIDDARTHAREDDY8]

The Label Checker workflow requires at least one label from its list (fix, feat, bug, etc.) — this PR currently has none, so the check fails on every run. Could a maintainer please add the fix label (this is a bug fix for #2954)? I do not have label permissions on this repo.

[Uburro]

this is PR identical with https://github.com/nebius/soperator/pull/2955/changes