Portable BGMN threading/parallelism + fit-aware intensity-order pruning

[lauren-walters]

Change 1 — Bound BGMN threads, keep worker parallelism (portable):

• New src/dara/hardware.py detects usable cores (SLURM env vars → os.sched_getaffinity → os.cpu_count).
• settings.py adds BGMN_N_THREADS (default 1, env DARA_BGMN_N_THREADS) and RAY_NUM_CPUS (default = detected cores, env DARA_RAY_NUM_CPUS).
• BGMN now runs 1 thread per refinement instead of 8, avoiding cores×8 OS-thread oversubscription. ray.init pins num_cpus=RAY_NUM_CPUS (SLURM-safe); the inner BGMN task uses .options(num_cpus=BGMN_N_THREADS); the outer orchestration task uses num_cpus=0 so it no longer pins a core while blocked, freeing all cores for BGMN refinements (concurrency ≈ cores/threads).
• generate_control_file default n_threads now sourced from BGMN_N_THREADS.

Change 2 — Don't discard a better-fitting branch on the intensity-order heuristic:

• Existing intensity-order / low-weight-fraction detection is unchanged.
• A flagged branch is now pruned only if it did NOT improve Rwp by at least 8% vs its parent (MATERIAL_RWP_IMPROVEMENT = 0.08), via the pure helper keep_branch_despite_intensity_order, guarded against missing/zero parent Rwp.
• New tests/test_search_tree.py covers the helper and integration paths.

Summary

Major changes:

• feature 1: ...
• fix 1: ...

Todos

If this is work in progress, what else needs to be done?

• feature 2: ...
• fix 2:

Checklist

• All existing tests pass.
• Tests have been added for any new features/fixes.
• Docstrings have been added in the Google docstring format.

Tip: Install pre-commit hooks to auto-check types and linting before every commit:

pip install -U pre-commit
pre-commit install