Blender Git Loki

We're now doing final conversions... float exrs to dpx amd png. Found a
couple of bugs in the new float support in sequencer;

- when effects have float buffer input, the effect strip itself only got
byte buffers assigned, which bypassed the float versions of effects
- the float version of cross effect used ints to fade... should be float
- tests if an effect should be recalculated didn't check for float yet
- display of float buffer in sequencer preview showed black when no
byte rect was in imbuf

Peter; we need this urgent, so I commit now.. check if this is according
to how you thought floatbuffers to work? thanks!