Complete recurrence matrix-exponentiation optimization and harden large-branch/update execution paths

[Copilot]

This addresses the remaining third optimization plan (matrix exponentiation) and aligns it with the already landed large-scale conditional/lazy-update improvements. The recurrence fast path now avoids unnecessary matrix materialization while preserving lazy-branching and pending-update semantics at scale.

• Matrix exponentiation fast path (recurrence evaluation)

  • Reworked recurrence evaluation in both scalar and vector formulas to apply exponentiation directly to the state vector instead of building M^k first.
  • Removes one full dense matrix accumulation path per random-access recurrence evaluation while keeping exponentiation-by-squaring behavior.

• Files updated

  • src/main/java/cod/range/formula/LinearRecurrenceFormula.java
  • src/main/java/cod/range/formula/VectorRecurrenceFormula.java

• Recurrence stress coverage expansion

  • Extended recurrence optimization demos to include deeper index reads and additional high-index outputs so matrix-exponentiation behavior is exercised beyond small checkpoints.
  • Added an additional simple constant-shift recurrence probe to broaden scalar recurrence coverage.

• Files updated

  • src/main/cod/demo/src/main/test/linearrecurrenceoptimization/LinearRecurrenceOptimization.cod
  • src/main/cod/demo/src/main/test/linearrecurrenceoptimization/VectorLinearRecurrenceOptimization.cod

private AutoStackingNumber[] applyMatrixPowerToVector(AutoStackingNumber[][] base, long exp, AutoStackingNumber[] vector) {
    AutoStackingNumber[] result = Arrays.copyOf(vector, vector.length);
    AutoStackingNumber[][] current = base;
    long e = exp;
    while (e > 0) {
        if ((e & 1L) == 1L) result = multiply(current, result);
        e >>= 1;
        if (e > 0) current = multiply(current, current);
    }
    return result;
}