FIX: properly handle zero epicentral distance without clamping r

[Dengda98]

• Remove the old practice of rewriting r=0 to 1e-5. Introduce GRT_IS_ZERO (threshold 1e-8) and evaluate near-field Bessel terms via exact kr→0 limits in k_integ, so displacement/derivatives remain well-defined on axis.
• At r=0, keep the two-step ZRT covariant split: synthesize the finite part of (1/r)∂_θ with ∂_r Green's functions, and replace u/r connection terms by ∂_r u in coord / stress / strain / rotation.
• Force azimuth to 0° (e_r→N, e_θ→E) and reject FIM/SAFIM when any distance is zero.