This commit implements on-the-fly filtering for KNN search, allowing the
search algorithm to continue exploring until k filtered candidates are
found, rather than finding k total candidates and then filtering them.

Key changes:
- Added `k` parameter to `searchBaseLayerST()` method to specify the target
  number of filtered results when filtering is enabled
- Added `use_filter` flag to determine when filtering mode is active
- Modified termination condition to continue searching until `k` filtered
  candidates are found when `use_filter` is true, instead of stopping after
  `ef` total candidates
- Updated exploration logic (`should_explore`) to continue exploring neighbors
  when filtering is enabled and fewer than `k` filtered candidates have been
  found, even if `ef` candidates have already been explored
- Added filtering check (`is_filtered`) before adding candidates to
  `top_candidates` priority queue, ensuring only filtered candidates are stored
- The filter callback (`BaseFilterFunctor* isIdAllowed`) is invoked for each
  candidate node via `(*isIdAllowed)(label)` to determine if it passes the filter
- Updated `searchKnn()` to pass the `k` parameter to `searchBaseLayerST()`
  when a filter is provided

The changes ensure that when filtering is enabled:
1. The search continues until `k` filtered candidates are found (if they exist)
2. Only candidates that pass the filter are added to `top_candidates`
3. The exploration continues even if `ef` candidates have been explored, as
   long as fewer than `k` filtered candidates have been found

This enables more accurate KNN search results when combined with attribute
filters, as the algorithm actively searches for filtered candidates rather
than relying on post-filtering which may return fewer than `k` results.