feat: vectorize CAM retrieval with NumPy and add multi-worker support

- Replace scalar hamming distance with NumPy bitwise_count for batch retrieval
- Add ThreadPoolExecutor-based multi-worker query parallelism
- Improve missing dataset error message with generation command hint
- Increase DEFAULT_MAX_QUERIES from 128 to 8192 for meaningful throughput tests

tests/test_prepare_cam_retrieval_dataset.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import numpy as np
 
 from scripts.prepare_cam_retrieval_dataset import (
+    DEFAULT_MAX_QUERIES,
     dataset_config,
     pack_bits_to_words_le,
     stratified_indices,
@@ -15,6 +16,10 @@ def test_dataset_config_resolves_cifar10_and_cifar100() -> None:
     assert dataset_config("cifar100") == ("uoft-cs/cifar100", "fine_label")
 
 
+def test_default_query_count_is_large_enough_for_software_throughput() -> None:
+    assert DEFAULT_MAX_QUERIES == 8192
+
+
 def test_dataset_config_rejects_unknown_dataset() -> None:
     try:
         dataset_config("mnist")