derive_entries_aware_chunk_name currently iterates over all entry points and calls has_bit for each one, every time an entries-aware chunk is created. If there can be many manual-code-splitting groups, this becomes O(num_chunks × num_entries). Consider deriving the set entry indices from bits (e.g., BitSet::index_of_one()) and/or caching a flattened entry list once, so chunk name generation scales better on large projects.

Originally posted by @Copilot in #8275 (comment)