Support writing files without knowing their size

[acidtonic]

Currently it seems you must know the size of the data going into the tar but in some cases where I'm porting IO to go into a tar, I don't know the file size and I'd prefer to avoid buffering.

Is is trivial to enhance this to keep updating the size as I write more data? I see some code to start writing null after size remaining is 0, perhaps I could simply modify that line to keep going?

Also would be cool if I could write multiple files at the same time (not multi-threaded though) by getting a descriptor when writing the file header. The API would then have you pass that descriptor to the write_data functions.

Appreciate the hard work :)