Reduce Memory Usage in Multi-Threaded INSERTs with Chunked Transform Collection

Reduce max memory usage of multi-threaded INSERTs by collecting chunks of multiple threads in a single transform. #61047 (Yarik Briukhovetskyi).