Affiliate pulse strategic improvements: trend, prevalence, cross-pulse

Three strategic gaps identified in HoP + Senior PM review of affiliate-onboarding-pulse (td-7d220b): 1. **Week-over-week trend signal** — Add a 4-week directional indicator to the Narrative. Currently only compares 7-day window vs 11-week baseline average. HoP needs 'is this better or worse than last week?' answered. Options: (a) compare to prior pulse output, (b) compute 4-week rolling in the script. 2. **Stall prevalence baseline** — Add one line in Narrative contextualizing whether this week's dominant stall type is normal or anomalous vs baseline distribution. Requires script 03 to also compute baseline stall-type distribution, not just signal-window stalls. 3. **Cross-pulse synthesis** — Flag overlap between affiliate stall list and referral stall list. Both pulses independently surface mixed merchants. One line noting how many of this week's affiliate stalls also appear in the referral stall list. Secondary (from Senior PM review): - Document baseline instability (90-day sliding window) as known limitation - Move 'program selection → setup conversion' to baseline-only metric at low volume - Add assertion in scripts to validate is_affiliate_campaign flag has non-zero count - Surface data caveats (survivorship bias, real-time AAM flag) in output footnotes