v9.0.0 - pragmastat

Now I have enough context to write the release notes.

Features

SpreadBounds estimator: Distribution-free confidence bounds on sample spread (median of absolute pairwise differences). Uses disjoint-pair sign-test inversion with randomized cutoffs for exact misclassification rate control. Implemented across all 7 languages (C#, Go, Kotlin, Python, R, Rust, TypeScript).
SignMargin toolkit function: Computes randomized exclusion margins for one-sample sign-test bounds using log-space Binomial CDF evaluation for numerical stability.

Fixed precision in erf and erf_inverse C# test approximation data (~1e-17 magnitude corrections)
Fixed precision in ratio-bounds/multiplic-10-30.json reference test data

Added mathematical study for SpreadBounds with design rationale, theoretical constraints, and algorithm derivation
Added SpreadBounds and SignMargin function specifications to the manual toolkit
Added SpreadBounds test suite documentation
Updated assumptions chapter with SpreadBounds sparity requirement and distribution-free bounds description

Added SpreadBounds Monte Carlo simulation to pragmastat-sim for validating observed vs. requested misrate
Added pragmastat-sim spread-bounds CLI subcommand
Updated reference test suite with 46 new JSON test files covering success cases, error handling, edge cases, algebraic properties, and unsorted input verification
Fixed fraction conversion and subscript handling in tools/src/math_conv.rs for Typst compatibility
Fixed table.header() syntax handling in tools/src/typst_parser.rs

Full Changelog: https://github.com/AndreyAkinshin/pragmastat/compare/v8.0.0...v9.0.0

Release notes generated by herald v1.0.4