Eou Preemption Analysis Sweep

Threshold sweep + wiring diagnostic for the synthesized-PREFLIGHT lever.

Separates two hypotheses for why Smart Turn's first 0.7-crossing fires after is_final: H1 — threshold too high: signal IS rising before is_final, just doesn't hit 0.7 in time. Lower threshold → earlier crossings. H2 — wiring issue: Smart Turn isn't invoked until late. Samples physically don't exist before is_final. No threshold change helps.

Per language returns:

• first_sample_advantage_ms_* — when the FIRST Smart Turn sample (any prob) fires relative to is_final. Positive = sample exists before is_final. Negative = no samples until after. This is the H2 diagnostic.
• max_prob_before_is_final_* — across the samples that DID land before is_final, what was the highest probability seen? Tells us whether lowering the threshold could possibly help.
• thresholds: per threshold (0.3, 0.4, 0.5, 0.6, 0.7), the crossed_pct and early_advantage_p50. The H1 diagnostic.