Add rev11: pre-generate MC RNG, preserve azimuth chunking, and add validator/benchmark

[nicklasorte]

Motivation

• Improve runtime by removing per-iteration rng(...) reseeding and eliminating repeated random draws inside the MC loop while preserving the aggregate max-azimuth output contract.
• Keep azimuth chunking as a memory/performance tuning knob rather than forcing full-azimuth expansion that can be memory-bound.
• Provide tooling to validate statistical equivalence and to find a practical AZI_CHUNK for different environments.

Description

• Added subchunk_agg_check_maxazi_rev11.m which seeds once with rng(rand_seed1) and pre-generates rand_pr_all, rand_eirp_all, and rand_clutter_all with dimensions [num_bs x num_mc_idx] and bounds [rel_min, rel_max] for use across MC iterations.
• Restored/kept azimuth chunking as a tunable parameter with AZI_CHUNK_DEFAULT=128 and optional override via varargin, and added a short comment explaining the memory/performance tradeoff.
• Preserved trusted aggregation semantics (dB→power via db2pow, sum across BS, power→dB via pow2db, then max over azimuth) and maintained nearest-point off-axis gain behavior.
• Reduced hot-path overhead by gating debug NaN checks behind DEBUG_CHECKS=false and tracking per-chunk maximums to avoid repeated full-vector max scans.
• Added validate_subchunk_agg_check_maxazi_rev10_rev11_statistical.m to run rev10 vs rev11 on identical inputs, report runtimes/speedup, compute distribution and upper-tail statistics, apply documented drift thresholds, and return a results struct with a PASS/FAIL flag.
• Added benchmark_subchunk_agg_check_maxazi_rev11_chunk_sweep.m to sweep chunk sizes (default [64 128 256 512]) across trials, print a summary table, and recommend the fastest observed AZI_CHUNK.

Testing

• Confirmed new files exist and were committed to the repository (subchunk_agg_check_maxazi_rev11.m, validate_subchunk_agg_check_maxazi_rev10_rev11_statistical.m, benchmark_subchunk_agg_check_maxazi_rev11_chunk_sweep.m) and the commit succeeded.
• Performed static verification of created file contents by listing/printing the new files in the worktree.
• Attempted to run automated runtime tests but MATLAB/Octave is not available in this environment, so the included validator and benchmark were not executed here and should be run in a proper MATLAB environment to validate runtime and statistical equivalence.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 36a8e200b0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Derive RNG stream from MC indices when pre-generating draws

This reseeds the generator with only rand_seed1 on every function call, so different sub_point_idx chunks regenerate the same random matrices whenever num_mc_idx is the same. In the chunked workflow (cell_sim_chunk_idx), that duplicates Monte Carlo realizations across chunks instead of producing unique draws per mc_iter, which can materially skew aggregate percentiles and reduce effective sample size. The pre-generation should incorporate sub_mc_idx/mc_iter identity (as prior revs did) or otherwise skip ahead in a deterministic way per chunk.

Useful? React with 👍 / 👎.