fix: correct kwargs key names in SlurmJob fluent setters

[prabindersinghh]

Summary

Fixes #10

The cpus_per_task(), memory(), and run_time_limit() fluent setter
methods wrote incorrect keys into self.kwargs, but
create_slurm_script() reads different keys. The mismatch meant all
three values were silently dropped — every job ran with default CPUs,
memory, and time limit regardless of what the user specified.

Setter method	Key written (before)	Key read by create_slurm_script()
cpus_per_task()	"cpus-per-task" (hyphen)	"cpus_per_task" (underscore)
memory()	"mem"	"memory"
run_time_limit()	"time"	"run_time_limit"

Changes

• metaflow_extensions/slurm_ext/plugins/slurm/slurm_job.py — 3 key renames
• test/unit/test_slurm_job.py — new file, 8 unit tests

Testing

8 new unit tests in test/unit/test_slurm_job.py:

Test	What it verifies
test_cpus_per_task_setter_writes_correct_key	setter writes correct key
test_memory_setter_writes_correct_key	setter writes correct key
test_run_time_limit_setter_writes_correct_key	setter writes correct key
test_create_slurm_script_includes_cpus_per_task_directive	script contains correct #SBATCH directive
test_create_slurm_script_includes_mem_directive	script contains correct #SBATCH directive
test_create_slurm_script_includes_time_directive	script contains correct #SBATCH directive
test_create_slurm_script_chained_fluent_api	all three directives present when chained
test_kwargs_constructor_path_still_works	existing constructor path unaffected

All 8 pass. Zero regressions.