The Trash Can undelete handling should not implement its own directory traversal handling to restore the attributes of files when "lfs trash unrm" is called on a directory.

The parallel directory processing implemented for parallel_find() should be used (and improved/adapted if needed) to avoid duplicate code, and allows focused optimizations for a single parallel directory traversal code instead of spreading development and bug fixing effort across multiple different implementations.

The current TCU directory traversal is BFS instead of DFS, but that can be addressed by a flag in struct find_param rather than having a fully duplicate implementation. Also, BFS vs. DFS is somewhat unclear/irrelevant when there are multiple threads performing directory traversal in parallel.