Fix ESDIRK659L2SA embedded coefficients so adaptive solves work (#3659)

[singhharsh1708]

Summary

Fixes #3659. ESDIRK659L2SA adaptive solves were running to MaxIters because the embedded btilde coefficients were inconsistent (sum(btilde) ≠ 0), breaking the error estimator while leaving the main order-6 method itself correct.

This PR replaces the embedded pair with a derived minimum-norm order-5 embedded solution satisfying the full order conditions for the fixed (A,c) tableau. The primary method (A, b) is unchanged; only the adaptive error estimate is corrected.

Verification

• Issue repro now returns retcode=Success with sane adaptive step counts
• Fixed-step convergence unchanged (order ≈ 6)
• Error scales monotonically with tolerance
• Stiff van der Pol (μ=1000) adaptive solves behave normally again
• tableau_consistency_tests.jl: 47/47 pass (ESDIRK659L2SA moved from @test_broken to @test)

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Add any other context about the problem here.

[singhharsh1708]

@ChrisRackauckas Do you have the paper's exact b̂ row to drop in instead? If so I'll use it verbatim as rationals

[ChrisRackauckas]

How does the paper state it? If it's long decimals, make them literal bigs like Verner

[singhharsh1708]

@ChrisRackauckas