Align exp_009 with F-1…F-7 remediation and wire PHY-effective SNR into AI scheduling

[Copilot]

This change shifts exp_009 from documenting known architectural flaws to verifying that the prioritized F-1…F-7 remediations are actually active in integrated behavior. It also closes the remaining F-1 gap in AI scheduling by ensuring PHY-enhanced SNR influences Q-bandit decisions.

• Experiment semantics updated (exp_009)

  • Replaced flaw-assertion flow with remediation checks for F-2/F-3/F-4/F-5/F-6/F-7.
  • Updated experiment narrative/output from “flaw surfaced” to “remediation verified.”
  • Kept the experiment focused on cross-layer evidence rather than static claims.

• F-1 closure in AI-native MAC scheduling

  • Q-bandit UE selection now evaluates each UE using scheduler-effective SNR (phy_effective_snr when present), not only baseline SNR.
  • This makes PHY gains (e.g., RIS/OTFS-informed estimates) actionable in AI scheduling decisions.

• Cross-layer signal path exercised in experiment logic

  • exp_009 now constructs PHY-enhanced channel state and feeds effective SNR into AI reward updates, so the closed-loop behavior is visible in the run.

// Q-bandit greedy selection now uses scheduler-effective SNR
.max_by(|(i, a), (j, b)| {
    let qa = self.q_value(*i, a.scheduler_snr());
    let qb = self.q_value(*j, b.scheduler_snr());
    qa.partial_cmp(&qb).unwrap_or(std::cmp::Ordering::Equal)
})