`GoldSelector`: Select a bunch of really similar data

**Context & Problem** After already isolated some `anchors` (for instance based on a GoldGreedyKCenterSelectionTool output or after defining centroids with clustering), one might want to select new samples based on their prioperties compared to the anchors. Let's add a selection tools returning different samples based on anchors! **Proposed Solution** 1. Add a base class GoldAnchorBasedSelectionTool (same as GoldSelectionTool but with an `anchors` attribute) 2. In `select`, consider k to be per anchor and do a loop over the anchors 3. add `_select_for_anchor` abstract method to be called in the loop of `select` 4. add a simple GoldClosestAnchorBasedSelectionTool: select the closest embeddings from each anchor Note: how would we get the information linking the selection and the anchor?