fix: align Trisomy 21 BMI upper bound

[pacharanero]

Summary

• Accept UK Trisomy 21 BMI calculations through the final source-table age of 18.92 years.
• Reject ages immediately above 18.92 with the existing LookupError path and corrected message.
• Add exact endpoint regression tests for both sexes using the final two source rows and math.nextafter() beyond the endpoint.

Source evidence

The canonical source workbook rcpch/growth-references/trisomy21/UKReference/LMS_Downs.xlsx contains BMI LMS values for both sexes at 18.83 and 18.92 years, then blank BMI cells from 19.00 onward. The packaged trisomy_21.json accurately contains those same final two rows.

The previous 18.82 guard did not correspond to a source-table age and made both final rows unreachable. This PR changes no reference values.

Follow-up

Issue #109 and roadmap item R25 track replacing duplicated age literals across all references with validated source/applicability metadata. That work must distinguish raw table extent from clinical applicability and hybrid-reference overlap policy.

Verification

• Focused boundary tests: 6 passed
• s/test --running -q: 145161 passed
• Ruff fatal-error checks on changed files: passed
• git diff --check: passed