Merging r60962 through r60963 from trunk into soc-2011-tomato

Merging r59178 through r60961 from trunk into soc-2011-tomato

Made branch closer to trunk

Merging r59173 through r59177 from trunk into soc-2011-tomato

Merging r59088 through r59172 from trunk into soc-2011-tomato

Merging r59029 through r59087 from trunk into soc-2011-tomato

Replace our own box AA filter with BLI_jitter-based

Using box filtering for AA makes image too much blurry,
so here's an attempt to use generic BLI_jitter routines
to do sampling.

It's expected to be exactly the same as opengl renderer
does, but i could have got something wrong.

But anyway, results seems to be a little bit less fuzzy.

Merging r59000 through r59028 from trunk into soc-2011-tomato

Fix crash when adding plane track in some cases

Need to re-evaluate plane after susing generic transform tools.

Delete Marker operator will now also tae plane markers into account

Smaller fixes:

- Point track delection missed checking whether track actually
belongs to the plane or not.

- Adding point track shall deselect all the planes.

- Fixed bug with restoring marker's flag after transformaiton.

Prevent plane tracks from being concave

Re-arrange AA sampling a bit

Apparently when calculating UV from upsampled
corners warped result seems too much doggy.

Now result seems to be exactly the same as it
was before AA changes (apart from AA-ed edges :)

Anti-aliasing for plane track image warping

Uses the same exact kernel size of 4x4 as for plane
track mask output.

Think this is not a bad way to go, difference in speed
using test file was only about 30% with much better
output result.

Made plane track transformable by generic transform operators

Pretty much straightforward change, apart form one trick needed
to preserve point tracks scaling nice:

- If TransData belong to point track, use it's individual center
when scaling (ignoring pivot point center).

This works much better for adjusting sizes of point tracks.

- If TransData belongs to plane track, use pivot point settings.

This is done via TransData's flag TD_INDIVIDUAL_SCALE which is
currently used by point tracks conversion code, but might be
used by other transform elements as well.

Plane mask anti-aliasing experiment

Currently uses hardcoded kernel size of 4, which means
node is actually 16 times slower now, but gives pretty
much nice anti-aliased results.

Fix incorrect plane tracks array initialization

It lead to crashes when trying to create plane track
from subset of existing point tracks.

Merging r58956 through r58999 from trunk into soc-2011-tomato

Make byte-float ocnversion threaded in compositor

In fact, there's no need to get float buffer at all,
conversion could be done in pixel processor level
after interpolation.

It might give slightly worse interpolation results
(which i'm not sure would be visible by eye) but
it gives more than 2x speedup on my laptop on node
setups used for warping image.

Mask points now could be parented to the plane track

MaskParent structure was extended by:
- type, which indicates whether it's a parent to point
or plane track. Types might be extended further.

- Original corners, which are used to obtain homography
used to deform point's coordinate.

Homoraphy is detecting between original corners position
(corners at the parenting time) and current plne corners,
and the it's being applied on mask point coordinate.

Some tricks with switching between mask and clip coords
is needed, but it's all pretty much straightforward in
the code.

Parenting happens on spline point level, not spline one.
This fits existing design, and it's not so big difference
for artist.

Parenting operator (Ctrl-P) might for sure be used to
parent points to the active plane track.

Additional change: added visualization of plane corners
orientation. Corner which corresponds to left bottom
image corner now has got two axis: red which corresponds
to X axis, and green one for Y axis. This way it's
easier to orient corners on the footage.