Fixed implementation of error reporting. Made all formats be compared with cuSPARSE (format and cuSPARSE compared to CPU, not between each other, that would require Benchmarks.h to be changed). Added name of mtx being tested to MetaDataColumns. Code reformatting.

Found potential mistake in SpMV/spmv.h where MatrixReader doesn't need to be called twice. Commiting to show in meeting.

Implemented rough version of result comparison. Implemented benchmark for comparison of TNL CSR and Cusparse on GPU. Edited log file formatting.

Implemented for the benchmark to write the output of MatrixReader into the log file. BUG: Every other error message added into the Benchmark doesn't have a '!' as a prefix in the log file.

Copied tnl-benchmark-spmv files and spmv.h from BLAS to SpMV. Deleted min/max size and stepFactor. Not working yet, backup purposes.

Closes #49

They are not suitable for more than 2 devices/execution types; their design
breaks the Open-Closed Principle. Instead, a type template "Self" was
created, which allows to change any template parameter.

The usage of algorithms such as MemoryOperations or Reduction is not
bound to a particular container. On the other hand, ArrayIO,
ArrayAssignment, VectorAssignment and StaticArrayAssignment are just
implementation details for the containers - moved into
TNL/Containers/detail/

Also moved ParallelFor, StaticFor, StaticVectorFor, TemplateStaticFor
into TNL/Algorithms/

It has nothing to do with devices.

Moved synchronization of smart pointers from Devices::Cuda into TNL::Pointers namespace as free functions

synchronizeDevice() was renamed to synchronizeSmartPointersOnDevice()
for clarity - there are many similarly named functions in CUDA (e.g.
cudaDeviceSynchronize()).

Moved (most of) static methods from TNL::Devices::Cuda as free functions into separate namespace TNL::Cuda

The class TNL::Devices::Cuda was too bloated, breaking the Single
Responsibility Principle. It should be used only for template
specializations and other things common to all devices.

The functions in MemoryHelpers.h are deprecated, smart pointers should
be used instead.

The functions in LaunchHelpers.h are temporary, more refactoring is
needed with respect to execution policies and custom launch parameters.

Fixes #46

Benchmarks can be easily profiled even without this parameter, so it was
just an unnecessary complication.

This is nicer because it more clearly separates data load, computation
and data store. Furthermore, it allows to use instances of std::plus,
std::logical_and, std::logical_or, etc. instead of custom lambda
functions.

It contained only methods for prefixSum and segmentedPrefixSum, which
were identical for Host and Cuda, so they can be easily implemented
directly in Vector and VectorView.