docs: evaluate ModeMath architecture

[Happypig375]

Summary

• add a reproducible architecture study comparing pinned ModeMath and CSharpMath revisions
• record the feature matrix, licensing constraints, code-size split, benchmark evidence, and adopt/prototype/reject decisions
• include a deterministic F# benchmark harness with exact package/source provenance and capture validation
• validate canonical aggregate names, groups, and sample identities so incomplete evidence cannot pass

Validation

• benchmark harness Release build
• isolated NuGet restore from the exact ModeMath package and verified SHA-256
• aggregate.ps1 -Check against the committed 44 metric rows
• negative checks for missing, duplicate, substituted, extra, and cross-library sample groups
• PowerShell syntax checks and git diff --check

Closes #273