fix(strategy): --data-rank help omits decaying-spectrum and misstates its default

[statxc]

PR kind

• fix
• feat

Summary

The --data-rank CLI help was inaccurate on two counts:

• Scope: it said "rank of generated matrices when --fill lowrank", but storage.generate uses data_rank for both --fill lowrank and --fill decaying-spectrum (r = data_rank or max(1, n // 32) in each branch).
• Default: it said default n//32, but the real default is max(1, n//32) — floored so small n (where n//32 == 0) never produces a rank-0 / all-zero matrix.

Empirically (pure NumPy, no GPU):

data_rank= 2 -> matrix_rank=2
data_rank= 8 -> matrix_rank=8
data_rank=20 -> matrix_rank=20   # for --fill decaying-spectrum

Fix

Correct the help to name both fills and the real max(1, n//32) default, and add tests pinning the contract.

Touches only strategy/cli.py and a new test — no protected paths.

Validation

uv run --extra test python -m pytest tests/test_data_rank_decaying_spectrum.py -q   # 3 passed
uv run --extra test python -m pytest tests/ strategy/tests/ eval/tests/ -q          # 284 passed, 24 skipped

Checklist

• CPU-safe validation passed.
• My commits do not include Co-authored-by footers for coding agents such as Cursor, Codex, Claude, Copilot, or similar tools.
• Not a feat PR — no scorecard required.