Stabilize CI test runs with deterministic environment settings

[harryswift01]

Summary

This PR resolves nondeterministic regression test failures by enforcing deterministic iteration, eliminating shared mutable state, and standardizing the CI execution environment. Parallel execution with pytest-xdist is preserved.

---

Changes

Deterministic data handling

• Enforce deterministic iteration over group mappings:
  • Replaced groups.keys() with sorted(groups.keys())
  • Standardized ordering when building group_id_to_index
• Ensure consistent floating-point accumulation:
  • Sort components before summation in results reporting
• Remove shared mutable state (aliasing):
  • Replace direct references with copies (list(...), .copy()) for force/torque accumulators

---

Deterministic CI environment

• Set thread-related environment variables to avoid nondeterministic BLAS/OpenMP behavior:
  • OMP_NUM_THREADS=1
  • MKL_NUM_THREADS=1
  • OPENBLAS_NUM_THREADS=1
  • NUMEXPR_NUM_THREADS=1
• Set PYTHONHASHSEED=0 to ensure stable dictionary hashing and ordering

---

Standardize test execution

• Keep pytest-xdist enabled (-n auto) for performance
• Use consistent distribution strategy (--dist=loadscope) across all jobs
• Align test execution flags across unit, regression, and coverage workflows

---

CI consistency improvements

• Apply identical environment configuration across all workflows (PR, daily, weekly)
• Ensure consistent behavior across operating systems and Python versions

---

Root cause

The regression failures were caused by a combination of:

• Non-deterministic dictionary iteration affecting group-to-index mappings
• Floating-point accumulation order differences
• Shared mutable state between data structures leading to cross-test contamination
• Threaded numerical libraries introducing small, inconsistent variations

These issues manifested as intermittent failures, typically in later tests or when running in parallel.

---

Impact

• Eliminates flaky regression test failures
• Guarantees deterministic results across runs, platforms, and Python versions
• Maintains parallel execution performance
• Improves reproducibility between local and CI environments

---

Validation

• Repeated regression test runs with parallel execution show consistent results
• Verified stability under varying PYTHONHASHSEED
• Added unit tests to ensure deterministic group indexing and prevent aliasing