perf+feat: Morse hot path, Metatomic keys (#215), ASE srun (#193)

[HaoZeke]

Summary

• perf(morse): From tmp_neb_profile/cachegrind.out (~60% Ir in Morse::force on NEB Morse): inline pair potential, inv-r force scale, pointer-based force accumulation.
• feat(metatomic) MAINT: Allow keys to be set by the user for Metatomic #215: [Metatomic] energy_output / energy_uncertainty_output for non-default model output names (variants still work).
• feat(ase-nwchem) MAINT: Handle srun in the ASE calculators better #193: [ASE_NWCHEM] mpi_launcher (mpirun default or srun).

Test plan

• CI green (towncrier, lints, Morse-using NEB tests)
• Optional: re-run cachegrind on same NEB Morse case for Ir delta

[github-actions]

Benchmark Results

Warning

1 benchmark(s) regressed

	Count
🔴 Regressed	1
🟢 Improved	1
⚪ Unchanged	6

Regressions

	Benchmark	Before	After	Ratio
🔴	bench_eonclient.TimeNEBMorsePt.time_neb	262±0ms	296±0ms	1.13x

Improvements

	Benchmark	Before	After	Ratio
🟢	bench_eonclient.TimePointMorsePt.time_point_evaluation	9.78±0ms	8.03±0ms	0.82x

6 unchanged benchmark(s)

Benchmark	Before	After	Ratio
bench_eonclient.TimeMinimizationLJCluster.peakmem_minimization_lbfgs	41.2M	41.2M	~1x
bench_eonclient.TimeMinimizationLJCluster.time_minimization_lbfgs	33.5±0ms	33.7±0ms	~1x
bench_eonclient.TimeNEBMorsePt.peakmem_neb	41M	41M	~1x
bench_eonclient.TimePointMorsePt.peakmem_point_evaluation	41.2M	41.1M	~1x
bench_eonclient.TimeSaddleSearchMorseDimer.peakmem_saddle_search_dimer	41M	40.9M	~1x
bench_eonclient.TimeSaddleSearchMorseDimer.time_saddle_search_dimer	68.2±0ms	63.4±0ms	~0.93x

Details

• Base: 0a40bb30
• Head: 7b9f0825
• Runner: ubuntu-22.04

Raw asv-spyglass output

All benchmarks:

| Change   | Before   | After    |   Ratio | Benchmark (Parameter)                                                  |
|----------|----------|----------|---------|------------------------------------------------------------------------|
|          | 41.2M    | 41.2M    |    1    | bench_eonclient.TimeMinimizationLJCluster.peakmem_minimization_lbfgs   |
|          | 33.5±0ms | 33.7±0ms |    1    | bench_eonclient.TimeMinimizationLJCluster.time_minimization_lbfgs      |
|          | 41M      | 41M      |    1    | bench_eonclient.TimeNEBMorsePt.peakmem_neb                             |
| +        | 262±0ms  | 296±0ms  |    1.13 | bench_eonclient.TimeNEBMorsePt.time_neb                                |
|          | 41.2M    | 41.1M    |    1    | bench_eonclient.TimePointMorsePt.peakmem_point_evaluation              |
| -        | 9.78±0ms | 8.03±0ms |    0.82 | bench_eonclient.TimePointMorsePt.time_point_evaluation                 |
|          | 41M      | 40.9M    |    1    | bench_eonclient.TimeSaddleSearchMorseDimer.peakmem_saddle_search_dimer |
|          | 68.2±0ms | 63.4±0ms |    0.93 | bench_eonclient.TimeSaddleSearchMorseDimer.time_saddle_search_dimer    |