Previously, the code returned a status and required the caller to mangle
this status with his own status. Now, the function takes the previous
status ass an argument and does the mangling itself.
Also contains fixes for all the callers to actually check the return
value - which is now rather trivial as it just requires passing through
the status variable.
For firefox-planet-gnome, 19135 times a gradient gets rendered using
only 10 different gradients. So we get a 100% hit rate in the cache.
Unfortunately, texture upload is not the biggest problem of this test,
as the performance increase is only moderate - at least on i965:
34.3s => 33.5s
