feat: Add JAX acceleration support to z_n_search

[temp-noob]

feat: Add JAX acceleration support to z_n_search

Adds optional JAX-accelerated backend for z_n_search via use_jax parameter.
The JAX implementation computes exact unbinned Z^2_n statistics directly from
event phases, complementing the existing numba-JIT'd binned approach.

Implementation details:

• JAX kernel using vmap for vectorized frequency/fdot grid search
• Conditional @jax.jit with static_argnums for nharm parameter
• Graceful fallback when JAX not installed (HAS_JAX flag)
• Compatible with existing API (1D and 2D search with fdots)

Added comprehensive test suite (16 tests):

• Correctness tests validating peak detection
• Comparison tests (JAX vs standard: >0.95 correlation)
• Benchmark tests showing ~19x speedup on 1D searches (CPU)

Performance on CPU: 1D (no fdot vector) searches gain ~19x speedup. Larger gains expected
on GPU-enabled JAX backends.

[temp-noob]

@matteobachetti, this is my first time contributing to the repo. Let me know if I can help out with more things or things which would be more crucial.