Initially, the timer for the ReadTotalTimeout started
counting on the beginning of the request. For large
uploads like LFS this could be problematic because the
timeout can be exceeded even before the server had a chance
to respond.

Changelog: fixed