Add serial chunked conformational map-reduce workflow

[harryswift01]

Summary

This PR implements the serial chunked conformational map-reduce workflow for dihedral conformational state analysis. The workflow now processes selected trajectory frames in deterministic chunks while still executing serially, preparing the conformational path for future Dask-backed execution without introducing distributed scheduling in this change.

Changes

Serial Chunked Conformational Workflow:

• Added a chunked conformational state-building workflow.
• Split conformational analysis into topology discovery, frame-chunk angle collection, global peak reduction, state assignment, and final state reduction.
• Preserved serial execution while introducing the same map-reduce structure needed for later parallel execution.

Dihedral Analysis Refactor:

• Replaced the monolithic dihedral analysis module with a domain-specific dihedrals package.
• Added dedicated modules for topology, angle observations, peak detection, state assignment, and numerical kernels.
• Reused cached chunk-local angle observations during state assignment to avoid rerunning dihedral calculations unnecessarily.

Regression-Safe State Reduction:

• Preserved the existing residue conformational state indexing behaviour to keep regression outputs unchanged.
• Kept deterministic chunk and molecule ordering during reduction.
• Added and updated unit tests for chunked angle collection, histogram reduction, peak detection, state assignment, and regression-preserving behaviour.

Impact

• Prepares conformational state analysis for future Dask-backed chunk execution.
• Improves maintainability by separating dihedral analysis into smaller domain-specific modules.
• Keeps current numerical regression behaviour unchanged.
• Reduces risk for the next parallelisation step by proving the chunked algorithm in serial first.