Stabilize CORE test suite by aligning deterministic test fixtures, paths, and aliases

[Copilot]

This PR scopes to CORE stabilization only: deterministic lumping/kernels/plots/config-path handling/linkage-Z conversion/macrostate assignment. It avoids deferred areas (stochastic workflows, GPCCA, RMSD extras, random frame export, PDB generation, least-moving-residue outputs).

• Deterministic test mapping + alias cleanup

  • Replaced stale tests/data/lumpings.yaml entries with current deterministic keys/values (t, kl, t_js, kl_js, js, gpcca; none/JS/ref semantics).
  • Updated stale dataset aliases in tests:
    • PDZ3_7 → PDZ3
    • aSyn_rdc_200ns → aSyn

• Config path assumptions corrected

  • Updated test invocations that assumed tests/data/<dataset>/config.yml to use tests/data/<dataset>/input/config.yml (including stochastic variants in input/).
  • Kept a single canonical config location (input/config.yml), without duplicating config files.

• CORE plot/test harness fixes

  • Corrected expected Z path resolution in plotting tests to remain dataset-scoped after config path normalization.
  • Ensured manual-inspection plot outputs create parent directories before writing.
  • Closed matplotlib figures around CLI test invocations to prevent cross-test figure/layout leakage.
  • Aligned PDZ3 macrotraj deterministic plot test to the available deterministic setup (KL/none).

• Deterministic runtime hardening + stale reference refresh

  • In MPP/MPP.py, guarded mean-feature assignment for states with no members to avoid NaN feature propagation in deterministic runs.
  • Regenerated stale deterministic reference:
    • tests/data/PDZ3/expected_output/kl/Z.npy
  • Added a minimal absolute tolerance for one deterministic scalar assertion to avoid false negatives from floating-point noise.

# Before: could emit NaN when a state has no assigned frames
self.mean_feature[i] = self.mean_feature_trajectory[self.trajectory == i].mean()

# After: only compute mean for populated states
state_mask = self.trajectory == i
if np.any(state_mask):
    self.mean_feature[i] = self.mean_feature_trajectory[state_mask].mean()