avoid arbitrary memory use in upload

[rogpeppe]

When uploading a file, the code was reading the entire
body into memory before making the request.
This caused out-of-memory errors when
uploading large files.

Now, if you provide a ReadSeeker, it will seek to
the start of the file rather than reading the whole
thing into memory.

It's tempting to change the API so that it requires
a ReadSeeker rather than a Reader, but that would break
the goose.v2 compatibility guarantee.

[axw]

please add a comment stating why you're setting req.Body, rather than passing it into NewRequest

[rogpeppe]

Good catch. It turns out that it wasn't necessary in fact, as NewRequest does a dynamic type check for io.ReadCloser.

I've also added some more tests that check the new logic - there were some significant gaps in coverage previously.

[rogpeppe]

$$merge$$

[mhilton]

$$merge$$