Add Slurm worker MPI mode and claim paging

[nkeilbart]

Summary

• add execution_config.slurm_worker_mpi_mode so the outer direct-mode worker-launch srun can opt into --mpi=none
• page resource-based job claiming across ready-job candidates so smaller jobs can backfill past an initial priority window
• document direct-mode multi-GPU / nested srun usage with --overlap, --nodes, and --ntasks

Validation

• cargo fmt -- --check
• cargo clippy --all --all-targets --all-features -- -D warnings
• dprint check

Known Follow-up

• resource backfill can let producer-style setup stages run farther ahead of GPU/postprocess stages than before, which can increase staged file count on HPC systems with strict file-count limits
• we still need an explicit backpressure / work-in-progress control for that workflow shape instead of relying on the old first-page claim behavior as an accidental throttle