n2plain-improved-scaling-on-multithread

[HarshitNagpal29]

Summary

This PR adds an opt-in reusable workspace for full N2Plain reconstruction, covering both hadron-collider and electron-positron algorithms.

The workspace reduces repeated per-event allocations by reusing reconstruction scratch arrays, jet storage, and clustering history while preserving the complete ClusterSequence semantics.

Changes

• Add N2PlainWorkspace with support for:
  • PseudoJet storage for pp algorithms;
  • EEJet storage for e⁺e⁻ algorithms.
• Add callback-scoped with_n2plain_reconstruction.
• Reuse:
  • reconstructed jet storage;
  • complete clustering history;
  • pp nearest-neighbour and distance arrays;
  • e⁺e⁻ StructArray reconstruction storage.
• Resize scratch arrays to the exact active event length.
• Support caller-owned reusable output buffers through:

  inclusive_jets!(output, clusterseq)
  

Compatibility

This is an additive API change. Existing users continue receiving independently owned results. High-throughput applications can opt into reusable storage and copy only the values that must be retained.