docs: explain VERIFICATION.md metrics in plain English (closes #35)

[DsThakurRawat]

Description

Adds docs/understanding_verification.md, a plain-English guide that demystifies the auto-generated VERIFICATION.md trust receipt.

Key Additions:

• Export Metadata Breakdown: Explains opset, denoising steps, chunk size, and model types.
• File Hashes: Details why SHA256 matters for integrity, reproducibility, and audit trails.
• Parity Metrics Explained: Clearly defines max_abs_diff, mean_abs_diff, pass/fail thresholds, and random seed reproducibility.
• Vertical-Specific Tolerances: A table providing acceptable max_abs_diff ranges based on use-case sensitivity:
  • Warehouse arms (< 1e-04)
  • Farm robotics (< 1e-03)
  • Aerial drones (< 1e-04)
  • Retail & Traffic AI (< 1e-03)
• Remediation Steps: What to do if validation fails (e.g., opset downgrade, precision check).

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Infrastructure / CI update

How Has This Been Tested?

• pytest tests/ passes locally
• reflex doctor sanity check
• Other (please specify): Verified formatting and cross-referenced with codebase/schema.

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have kept the PR scoped to a single concern (one concern per PR)

[rylinjames]

Thanks @DsThakurRawat — the teaching structure of this guide (When is it created → Section-by-section → Key metrics → Vertical interpretation → Recovery) is exactly right, and the fixture numbers in your parity example look like real output from a reflex validate run on pi0.5 (the aggregate 2.384e-07 matches the README's pi0.5 first-action 2.38e-07 within 1 ULP — that's a real measurement, not a placeholder). Credit preserved via Co-Authored-By on the commit in the replacement PR.

What changed in the replacement, narrower than I'd initially planned:

• Refreshed only the Reflex version field (0.2.1 → 0.9.6 — that label was the only thing that didn't line up with current state). All your fixture numbers retained verbatim.
• Added a cross-reference to the actual parity ledger so readers can sanity-check their own runs against the shipped numbers (SmolVLA 5.96e-07, pi0 2.09e-07, pi0.5 2.38e-07, GR00T 8.34e-07).
• Added the Total: N files, X size line above the Files table — the renderer (src/reflex/verification_report.py) emits it.
• Renamed file to verification.md for consistency with sibling docs (eval.md, embodiment_schema.md).
• Replaced the cross-ref to TROUBLESHOOTING.md (doesn't exist on main yet — still in PR docs: compile TROUBLESHOOTING.md with CUDA and drone errors (closes #63) #129) with a reflex doctor flow.

Closing in favor of the replacement PR. Thanks again — this was the kind of doc that needed grounding more than rewriting.