workspace.add_file without incremental refcounts

I don’t know if this is intentional: When looping over large workspaces and adding files to them, the workspace.mets._file_by_id dict keeps growing and thus leaks all the new OcrdFile and METS file element etree references. This can create extreme memory overhead and slow down eventually (because each new file has to be compared to ever more existing IDs).

Perhaps one can at least try to sever the references to the file etrees?

But an opt-out to the general _file_by_id dict mechanism is probably best seen as part of #416, right?

[kba]

I noticed that as well when removing the trivial caching. On my TODO list.

[bertsky]

BTW This is a common cause of inefficiency. Not just when growing workspaces by adding more and more files. Any processor on a large workspace has to pay the penalty before it can start.