Defensive validation, numerical robustness, golub_welsch Result API (…

…v0.2.0) (#13)

* Defensive input validation, numerical robustness, and golub_welsch Result API (v0.2.0)

Address 12 deferred codebase review items across 5 phases:
- Reject ±Inf (not just NaN) in tanh-sinh, oscillatory, Cauchy PV, cubature
- Replace partial_cmp().unwrap() with unwrap_or(Ordering::Equal)
- Promote debug_assert to assert in CubatureRule::new
- Split Lobatto n<2 error into ZeroOrder (n=0) vs InvalidInput (n=1)
- Clamp global_error to non-negative in adaptive cubature
- Add dimension cap (d≤30) for adaptive cubature
- Clamp Newton iterates in Lobatto interior node computation
- Fix tanh-sinh non-convergence error: use last two estimates instead of fabricated tol*10
- Fix ln_gamma reflection: .sin().ln() → .sin().abs().ln()
- Document QUADPACK error simplification in gauss_kronrod
- Make golub_welsch return Result, propagate through all compute_* functions
- Bump version to 0.2.0, update CHANGELOG, SECURITY, README

* Run cargo fmt