Fix O(T²) backward trajectory, extend backward sampling to single-cluster models

[Copilot]

Three issues remained open from the initial CPF-IBS review. This PR addresses all three.

Changes

• O(T²) → O(T) trajectory assembly (src/particle.cpp)
  assemble_backward_trajectory was prepending to Armadillo matrices via insert_cols/insert_rows on every iteration, shifting all existing data each time. Replaced with pre-allocated matrices (set_size(n_items, T+1) etc.) written directly at column index t in the backward loop.

• Backward sampling extended to single-cluster models (src/rejuvenate.cpp)
  assemble_backward_trajectory was only reachable inside the if (prior.n_clusters > 1) block, so backward_sampling = TRUE was silently a no-op for single-cluster models and the corresponding tests were not exercising the new code path. Two changes:

  • stored_weights is now also transferred during MH acceptance so the weights are always current.
  • Added else if (options.backward_sampling) after the multi-cluster block to call assemble_backward_trajectory for single-cluster models.

  // Before: backward sampling unreachable for n_clusters == 1
  if (prior.n_clusters > 1) {
    ...
    if (options.backward_sampling) this->assemble_backward_trajectory(T, resampler);
  }
  
  // After
  if (prior.n_clusters > 1) {
    ...
    if (options.backward_sampling) this->assemble_backward_trajectory(T, resampler);
    else sample_particle_filter();
  } else if (options.backward_sampling) {
    this->assemble_backward_trajectory(T, resampler);
  }

• Duplicate alpha_hat assignment removed (tests/testthat/test-compute_sequentially_complete.R)
  Redundant back-to-back identical assignment that overwrote itself.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.