Adds WETW distance support to MESSI. Changes are additive - DTW and L2 untouched. Four files changed. See commit message for full breakdown.

[arihant1006]

Integrates Weighted Euclidean with Time Warping (WETW) as a third distance type alongside the existing L2_SQUARED and DTW modes.

Changes:

• DistanceComputer.hpp: add WETW = 2 to DistanceType enum
• iSAXIndex.hpp: add min_weight_vec field to isax_node for per-subtree per-dimension minimum weight vectors (filled bottom-up post index build)
• Messi.hpp: WETW fields in MESSI_workerdata, new private methods and data members, public setWeightData/setQueryWeights/computeMinWeights API
• Messi.cpp: four new functions mirroring the DTW pipeline: insert_tree_node_m_hybridpqueue_WETW (node-level LB traversal)
  calculate_node_WETW_inmemory (leaf-level 2-cascade filter)
  MESSI_topk_search_worker_WETW (parallel worker thread)
  MESSI_search_topk_WETW + searchIndexWETW (query entry points)
  Plus compute_node_min_weights + computeMinWeights for the bottom-up
  per-node min weight vector computation.

Lower bounds used:
Node level: min_j(w_q[j] + node->min_weight_vec[j]) * MINDIST_iSAX
Series level: ||(w_q + w_min_db) ⊙ (q - c)|| (per-dimension)
True dist: ||(w_q + w_c) ⊙ (q - c)|| (exact WETW)

Usage after this patch:
Messi messi(DistanceType::WETW);
messi.buildIndex(data_source); // index built on embeddings
messi.setWeightData(db_w); // per-series weight vectors [N*dim]
messi.computeMinWeights(); // fills node vectors bottom-up
messi.setQueryWeights(query_w); // query weight vector [dim]
messi.searchIndex(q, n, k, I, D);