This commit does too much. The first goal was to change our approach to saving new buffers so that we don't need to construct a File for an entry that doesn't exist. Rather than doing that, we call `Worktree::save_buffer_as` with the buffer handle, the path, and the contents. This then saves the buffer and returns a handle to a `File` that references an entry that actually exists. I needed to do this so that we can store an entry id on `File`.
In the process, I noticed intermittent test failures on `test_rescan_simple`, so I made some changes required to fix those related to our reuse of existing ids. Our previous approach of removing a path when inserting a new entry was broken, because of the recursive nature of `remove_path`. Instead, I simply recycle the id of an existing worktree entry with the same path if one is present, then allow it to be replaced.
Nathan Sobo
b7a4393f29
