(see also r8895, r8887 and #2043)

#1349 tbb::mutex is moved to outer scope to be sure that its initialization is completed before the parallel loop

lot's of changes; nonfree & photo modules added; SIFT & SURF -> nonfree module; Inpainting -> photo; refactored features2d (ORB is still failing tests), optimized brute-force matcher and made it non-template.

Small change in the SURF algorithm to be sure that without TBB the template parallel_for is not used
(the reason: the parallelization in the SURF algorithm conflicts with ThreadingFramework, which is a temporary substitute of TBB for ARMs)

replaced alloca() (a.k.a. cvStackAlloc) with AutoBuffer or vector() everywhere. cvStackAlloc() is still defined, but we do not need alloca() anymore to compile and run OpenCV (fixes #889 and may be some others)

temporarily turn off TBB threading when computing SURF descriptors (as the parallel version seems to produce different descriptors on each run)

integrated multi-threaded version of SURF (thanks to imahon and yvo2m for the patch; see ticket #275)