matrix: masking, crypto, core::simd column, staleness corrections, audits

Appends sections N-S to agnostic-surface-cpu-matrix.md, built from
read-only per-function audits (file:line cited throughout):

- N  masking ops per realization (compare/algebra/ternlog/fold, masked
     reductions and group-by — 19 scalar by design, with reasons), gaps
     and parity coverage; post-#323 rows for popcount/Hamming
- O  crypto lane (U32x16/U64x8 ARX per backend, in-tree consumers, gaps)
- P  core::simd counterpart of every simd_nightly method
     (native / composed / scalar-fallback / missing) and the gaps that
     are upstream candidates for portable-simd
- Q  staleness corrections to sections A-L (runtime-dispatch shipped,
     wasm backend exists, CI matrix, default config flip, ...)
- R  AMX cell rule: encoded-only / executed / executed-under-contention
- S  consumer audits: Belichtungsmesser (cascade) and Fisher-z

Sections A-L are left as written; Q records what in them is stale.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019HnekoM1EidTwQLS3oFVFm

Author: claude