Add cause-aware batch job requeue limits

[Yash0270]

Summary

Add cause-aware requeue limits for batch jobs so node failures and preemption/operator requeues can be bounded independently from MaxBatchRequeue.

• MaxNodeFailRequeue defaults to 5 and bounds node-failure requeues.
• MaxPreemptRequeue defaults to 0 (unlimited) and bounds preemption/operator requeues.
• MaxBatchRequeue keeps its existing launch/prolog failure behavior.
• Jobs that exceed a configured cause limit are held with JobHeldMaxNodeFailRequeue or JobHeldMaxPreemptRequeue; scontrol release resets the counters.

This change was reviewed internally at Meta in Yash0270/slurm#4. We had originally planned to send a direct patch, but given the recent changes to CONTRIBUTING.md, we are opening it as a pull request here instead.

Target Release

Targets master / the next stable Slurm release because this adds slurm.conf parameters plus protocol/state-file guarded job fields.

Issue

Reported in SchedMD support ticket: https://support.schedmd.com/show_bug.cgi?id=24298

Implementation Notes

• Adds per-job node-failure and preemption requeue counters.
• Keeps the counters controller-internal and packs them only under the 26.11 protocol/state guard.
• Updates config parsing/reporting, data parser coverage, Perl config bindings, reason-code docs, slurm.conf(5), and CHANGELOG/slurm-26.11.md.
• Leaves generated build-system files untouched.

Memory Impact

Two uint32_t counters per job = 8 bytes per job (~0.74% of the ~1,074-byte job_record struct).

Cluster size	Existing job_record memory	Our addition
1,000 jobs	~1 MB	8 KB
10,000 jobs	~10 MB	78 KB
100,000 jobs	~102 MB	781 KB
1,000,000 jobs	~1 GB	7.6 MB

The counters are controller-internal: packed under the for_state guard (state-save files only, not in scontrol show job / squeue RPC responses), so there is no network/protocol overhead.

Testing

• Whitespace/sanity: git diff --check origin/master...HEAD.
• Added ATF coverage in testsuite/python/tests/test_105_11.py for node-failure holds, preemption/operator holds, user requeue and requeuehold exemptions, counter independence, scontrol release counter reset, and MaxPreemptRequeue=0 unlimited mode.
• Functional testing: a Docker-based two-node e2e harness (Yash0270/slurm#5) exercised the same paths against a live cluster with real slurmctld / slurmd daemons; result: 9/9 PASS.

Contribution Notes

• Single logical commit with Signed-off-by and Changelog: trailers per CONTRIBUTING.md.
• No changes to generated build-system files (Makefile.in, configure).

[Yash0270]

testsuite/python/tests/test_105_11.py merge conflict

It's due to new code landed in master. Once PR is reviewed, we can rename the test to the latest test number available.