Add memory-aware two-level MC chunk planner (rev2) and grouped-save parfor wrapper (rev9)

[nicklasorte]

Motivation

• Reduce filesystem load and keep per-worker memory safe for very large Monte Carlo runs by separating compute chunk sizing from save-file grouping.
• Avoid the previous one-file-per-compute-chunk pattern that caused slowdown when the saved chunk count grew beyond ~200.
• Preserve restart/recovery behavior and keep the design safe for parfor by ensuring unique output files per grouped save chunk.

Description

• Added dynamic_mc_chunks_rev2.m implementing a two-level planner that computes a memory-safe compute_chunk_size from worker_memory_mb, target_memory_utilization, num_bs, and num_sim_azi using a conservative safety multiplier (4x), then groups compute subchunks into num_save_chunks capped by max_saved_chunks (preferred cap 128).
• Added parfor_randchunk_aggcheck_rev9_claude.m which processes one save chunk at a time, iterates its memory-safe compute subchunks via subchunk_agg_check_rev8, accumulates results in memory, and writes a single grouped .mat file per save chunk using a temp-file-then-movefile pattern and retry logic to reduce partial-write/corruption risk.
• Added validate_grouped_chunk_strategy_rev1.m which provides automated integrity checks and an optional numerical A/B path that reconstructs baseline (per-compute) vs grouped-save outputs and verifies size equality, NaN pattern equality, max absolute difference, randomized-order invariance, and restart-skip behavior.
• The design preserves existing .m files unchanged, returns a structured chunk_plan (mappings, ranges, randomized save order, and derived sizing info), and enforces parfor-safety by ensuring no concurrent writes to the same file.

Testing

• Performed automated static checks by creating and inspecting the new files dynamic_mc_chunks_rev2.m, parfor_randchunk_aggcheck_rev9_claude.m, and validate_grouped_chunk_strategy_rev1.m to verify the expected interfaces and required fields are present.
• Attempted to run a runtime check via octave --version but the Octave/MATLAB runtime was not available in this environment, so no numerical tests were executed here.
• The new validate_grouped_chunk_strategy_rev1 implements the automated numerical checks (baseline vs grouped assembly, randomized-order invariance, and restart-skip detection) and can be run in a MATLAB/Octave environment against representative inputs to confirm numerical equivalence and restart behavior.

---

Codex Task