Add support for GPU-accelerated operations via Boost.Compute

Add support for a new (arbitrary-rank) tensor type and associated operations.

[4024] improve performance of inplace_solve

[6511] Division by scalar should use enable_if<>

[7297] Make the free functions 'num_columns' and 'num_rows' support the uBLAS traits system and better work with expression types

[7296] fixes and improvements to test utility functions

[7363] fixed coordinate_matrix::sort() for gcc 4.7 and others

[3968] fixed coordinate_matrix sort problem on MSVC10

[3539] changed computation of norm_inf for complex types to match mathematical definition.
Note: This might cause a performance drop because now std::abs(z) is called for each vector element. The old implementation used std::max(std::abs(real(z)),std::abs(imag(z)). Further norm_inf and norm_1 will now return the same values for complex vector.

[3501] Moved free functions in concepts.hpp into anonymous namespace.

Move semantics of vector/matrix container assignments have been implemented. They can be enabled by setting BOOST_UBLAS_MOVE_SEMANTICS. More details are on the preprocessor options page.

Introduce new free functions. See [3449], the new tests in libs/numeric/ublas/test and the inline documentation of the files in boost/numeric/ublas/operation/.

[3293] Fix resizing problem in identity_matrix

[3499] Add DefaultConstructible to concept checks