fix: raise MAX_RATE_DOMAIN from 30 to 100 per 1,000 students

The previous cap of 30 was exceeded by over 60% of districts — mostly sparse
groups with small enrollment cells whose Agresti-Coull upper bounds genuinely
extend that far. Analysis across all 51 states (16,279 districts) showed the
median max x-value is already ~56 per 1,000; only truly degenerate cases like a
single predicted arrest in a four-student cell (~250/1000) need capping.

A cap of 100 still guards against these outliers while letting realistic data
drive the axis for the vast majority of districts. The existing 'clipped' flag
and note mechanism remain unchanged — they activate only when extreme values are
encountered.

AGENTS.md

@@ -170,8 +170,8 @@ time, the port does not).
   posterior predictive is drawn as discrete mass, because it *is* discrete; a
   Gaussian KDE over four achievable values renders as a lumpy smear that reads
   as a rendering bug.
-- **`src/utils/rateDomain.js`** — `MAX_RATE_DOMAIN = 30` caps the axis so a
-  four-student cell can't squash every other curve. It reports `clipped` so the
+- **`src/utils/rateDomain.js`** — `MAX_RATE_DOMAIN = 100` caps the axis so a
+  four-student cell can't squash every other curve (up from 30, which was exceeded by over 60% of districts). It reports `clipped` so the
   chart says so instead of silently cropping.
 
 ## Common Pitfalls & Gotchas