Cache customised residue axes topology for frame covariance

[harryswift01]

Summary

This PR improves customised residue covariance performance by caching static residue axes-topology metadata before frame execution. The frame covariance loop can now reuse cached residue topology instead of repeatedly rediscovering the same heavy-atom, UA-mass, and neighbouring-bond information with MDAnalysis selections.

Changes

Cached residue axes topology:

• Added residue-level topology metadata to the existing axes-topology cache.
• Cached frame-independent residue data, including residue heavy atom indices, residue UA masses, and neighbouring-residue bond status.
• Stored cached residue topology in shared_data["axes_topology"].

Cached residue axes execution path:

• Added a cached residue axes method in AxesCalculator.
• Updated FrameCovarianceNode to use cached residue topology when available.
• Preserved the existing get_residue_axes(...) path as a fallback.
• Kept frame-dependent values such as positions, centres of mass, axes, moments of inertia, forces, torques, and covariance matrices computed per frame.

Tests:

• Added unit coverage for cached residue topology construction.
• Added unit coverage for cached residue axes calculations.
• Updated covariance-node tests to cover the cached residue path and fallback behaviour.

Impact

• Reduces repeated MDAnalysis selection work inside the customised residue covariance path.
• Improves performance for workflows using customised axes.
• Keeps the optimisation scoped to residue axes topology only.
• Preserves existing force, torque, covariance, entropy, Dask, and scheduling behaviour.
• Maintains fallback behaviour when cached residue topology is unavailable.

[jimboid]

More MDA performance optimisations. PR looks like it has all of the necessary changes to accompany the performance improvement based alterations.