Guess the mimetype of uploaded files when they're unknown.

Browsers can send an empty mimetype when they don't know what kind of
file we're working with. This triggers a code path that fails, and even
with that fixed, makes a number of files useless. For example, text
files without .txt, or images masquerading as .bin files, or shell
scripts.

Now, if we're not told of a good file type, and `file` exists, we'll use
that to determine the type of file. This captures a number of file types
we otherwise wouldn't be told about.

Uploaded some files I had sitting around. AUTHORS, some .sh files, renamed PNGs.
The thumbnails ended up correct.