docs: add Max's critique of April 24 experiment scope pivot

[max-toscano]

Title: docs: Max's review of April 24 experiment scope pivot

Body:

Summary

• Side-by-side review of Josh's proposed experiment cuts (commit 3ded7eb) against my independent analysis
• Identifies where we agree (H100 drop, Exp 4/5 cut, B7 Option 2) and where we diverge (Experiment 3, 3B on RTX 3060, venue, DRAFT-HOLD visibility)
• Risk analysis for both plans with specific failure modes
• Two diagnostic tests that should run before committing to either strategy: 3B hardware validation and OEI alpha-sweep

Key disagreements

1. Experiment 3: Josh cuts entirely. I say keep a simplified 3-point VRAM measurement (30 min to run) so the Proposition has empirical backing.
2. Llama 3.2 3B on RTX 3060: Josh assumes it fits. Nobody has tested it. One 5-minute command settles this.
3. Venue: Josh targets Main Track (May 6). I recommend NeurIPS Workshop (July-Aug) given the current evidence base.
4. DRAFT-HOLD blocks: Josh leaves as internal LaTeX comments. I say convert to visible Limitations or fix the claims before submission.

Action requested

Review the critique, then let's run the two diagnostic tests (3B gate + OEI sweep) before finalizing the strategy.