In some cases canceling render jobs can take a long time. Eg when
using database layers over a sloooooow connection, canceling a job
can be blocked by minutes while waiting for the first batch of feature
fetching to finish. (Since eg postgres features are fetched in batches
of 2000 with no opportunity to abort mid-way through this).
This meant that while the first render allows the GUI to remain
responsive, any subsequent render operations which occured before
the first render completes locks up the whole ui until the first
render can finish cancellation.
With this change, the render cancelation happens with blocking.
It means that you can pan and zoom around a map over of slow
connection without any ui locks.
