[awennersteen]

Description of Change

Adds a CTest suite under plugins/tests/spank_qrmi/ that runs the plugin hooks the way slurmstepd does for a batch step (slurm_spank_init -> slurm_spank_task_init -> slurm_spank_exit), without a Slurm cluster or a QPU.

• The test #includes spank_qrmi.c and provides fake spank_*/slurm_* and qrmi_* functions against the real slurm/spank.h and qrmi.h, so it links neither libslurm nor libqrmi. Each case records calls and checks the resulting job environment.
• Each case runs in its own process with a 5 s alarm, so a hang shows up as a failure.
• 17 cases: no --qpu, task step skipped, missing/failing config, one or several resources, unknown resources (alone and in a list), inaccessible resource, failing acquire, --env: plugin arguments (with and without =), copying {name}_QRMI_* from the job environment (including entries without =), token export and release with the matching deallocator, a fresh token replacing one inherited from the job environment, and RUST_LOG from SRUN_DEBUG.
• CI runs the suite after the first spank_qrmi build, which fetches qrmi.h.

One case is WILL_FAIL because it hits a bug on main; whichever of this PR and #204 merges second should drop it from the list in CMakeLists.txt:

• fresh_token_replaces_job_env_token -> Keep credentials out of spank_qrmi logs #204

(unknown_resource_in_list was WILL_FAIL until #209 merged.)

Testing

• ctest in Ubuntu 24.04 (libslurm 23.11) against main 1c2eb64: 17/17 pass (15 pass, 2 expected failures).
• Against Fail the job when a --qpu resource is not in the config #209 and Keep credentials out of spank_qrmi logs #204 respectively, the matching WILL_FAIL case passes.
• After rebasing on main 3455388 (adds a spank_job_control_setenv fake for spank_qrmi: expose --qpu value to slurmctld as QRMI_QPU job control e… #211), built against qrmi main: 17/17 pass.

Prepared by Claude, not yet reviewed by @awennersteen.