This continues the split of Device into Execution and Allocator. The
execution types in TNL/Execution are: Sequential, OpenMP and Cuda
(Execution::OpenMP is instead of Devices::Host).

TODO:

- smart pointers: replace Device with Allocator (methods getData() and
  modifyData() should be removed, instead there should be getHostData()
  and getImageData() in both const and non-const variants)
- serialization: use a placeholder string (like "any") because data from files should be loadable with any Executor or Allocator
- revise BuildConfigTags for problem-solvers
- compatibility of Executors with Allocators
- dynamic execution policy - to specify runtime parameters for a
  specific (parallel) algorithm
   - implementation:
      some hierarchy of class templates which have the static execution policy as a template parameter
      specific classes for certain algorithms (like Reduction or PrefixSum)
      e.g. `DefaultExecutionParameters<CUDA>` → `ReductionExecutionParameters<CUDA>`
                                              → `PrefixSumExecutionParameters<CUDA>`
      most algorithms should use `DefaultExecutionParameters<DeviceType>`
   - then:
      - extend tests:
         ParallelFor: achieve full coverage with small array size
         finishing reduction and multireduction on host/GPU
         prefix-sum: specify suitable maxGridSize, blockSize, elementsInBlock and decrease VECTOR_TEST_SIZE
      - cuda reduction: profiling + probably change "finish" to launch only 1 block of threads
         - try zero-copy buffer on the host instead of CudaReductionBuffer
      - try using `ParallelFor` with a specific block size in LBM
      - custom kernel launch configuration for traversers

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.

Serialization in TNL::File: File::save and File::load are specialized by Allocator instead of Device

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/

This will be necessary to avoid code bloat with more than 2 devices
(execution types).

- cudaMemcpy is slower than our ParallelFor kernel for CUDA
- use std::copy and std::equal instead of memcpy and memcmp, but only as
  sequential fallbacks
- use parallel algorithms for containsValue and containsOnlyValue (again
  with sequential fallbacks)

It has nothing to do with devices.

Also set the buffer size to 1 MiB, because larger buffer size slows down
memory copies significantly (e.g. MeshTest would take about 10x longer).

Addresses #26

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

The implementation for std::string (which is a base class of
TNL::String) is perfectly sufficient.

Fixes #11

StaticArrayAssignment expects the arguments passed as reference.

Tutorials

See merge request !41

Added command line parameter --tests-jobs to build script to set number of processes for unit tests.