ANYsolver 0.4.7 — accepted performance and runtime fixes

ANYsolver 0.4.7 packages the accepted static, spectral, and nonlinear performance work since 0.4.6, plus runtime failure-status corrections.

• Representative complete static analysis improved 3.98× and retained load-only analysis 19.85× on the declared workload.
• Retained medium-panel spectral analysis improved 11.52% in the seven-pair gate, with identical recorded modes and residuals.
• The representative nonlinear gate improved its four non-easy cases by a 12.44% median with passing physical comparisons. The follower-shell case was 0.96% slower.
• Force-control reaction reuse, fixed dead-load preparation, trial-tangent promotion checks, and solver-work diagnostics are included.
• The generated-geometry runtime facade preserves constraints and nonlinear failure details for locally edited models.

The Armijo line search is optional and remains experimental. Its separate convergence promotion decision was NO-GO; no convergence benefit is claimed. Element equations, physical tolerances, defaults, and coordinated dependency bounds are unchanged.

The full PR and post-merge CI, installed-package gate, and TestPyPI rehearsal bind this release to the reviewed source. See PR #58, release details, and the TestPyPI staged package.