feat: add parallel distance computation and vectorized pipeline

Rewrite the distance computation engine from scratch on top of v0.3.5:

- Vectorized kNN distances using NumPy broadcasting with chunked
  processing for memory efficiency and progress bar support
- Add n_jobs parameter for cross-cluster multiprocessing via
  concurrent.futures (n_jobs=-1 uses all cores)
- Restructure Numba path with non-generator kernels that support
  numba.prange for thread-level parallelism
- Optional scipy.spatial.distance.cdist and scipy.special.erf
  acceleration when scipy is available
- Vectorize _standard_distances, _prob_distances, and
  _norm_prob_outlier_factor pipeline methods
- Fully backward-compatible: all existing API calls work unchanged

Closes #36

Made-with: Cursor

Author: vc1492a