Optimize ellipk and ellipe with binary tree branching

Replace O(n) linear if-elseif chain with O(log n) binary tree branching
for polynomial table selection. Key optimizations:

- Add _ellipk_core and _ellipe_core with binary tree structure
- Use integer index (idx = unsafe_trunc(Int, m*10)) for faster icmp vs fcmp
- Simplify wrapper functions to handle only negative m and special cases
- Add special handling for extremely negative m where x rounds to 1.0

Performance improvement:
- ellipk: 1.33x average speedup (10.5ns -> 7.9ns)
- ellipe: 1.46x average speedup (12.0ns -> 8.2ns)
- Largest gains for m in [0.65, 0.875]: 1.5x-1.85x speedup

All polynomial coefficients unchanged - only branching structure optimized.

Author: mgyoo86