This is necessary for local indexing of DistributedNDArray.

It works for any value, device and index types, but the permutations of
both arrays must be the same and both arrays have to be contiguous.

- fixed executors for operations: use inverse permutation when calling
  the wrapped lambda function
- custom internal region can be specified with custom begins/ends
  multiindices

NDArray: moved SizesHolderStaticSizePrinter and SizesHolderSizePrinter into the implementation namespace

Fixed expression templates

Brief summary:

- fixed `DistributedExpressionTemplates`
- many fixes and simplifications in `ExpressionTemplates` and `StaticExpressionTemplates`
- extended tests (unified for `VectorExpressions`, `VectorViewExpressions`, `StaticVectorExpressions`, `DistributedVectorExpressions`, `DistributedVectorViewExpressions`)
- functions `maxNorm`, `l1Norm` and `l2Norm` - convenient aliases for `max(abs(...))`, `lpNorm(..., 1)` and `lpNorm(..., 2)`
- a `cast` function which makes something like this possible:
  `float s = TNL::sum(cast<float>(vector))`, where vector elements may be `int`, `double`, etc.
- fixed result types in expression templates and vertical operations
- removed `addVector`, `addVectors`, `addElement`, `scalarProduct` and `sum` methods from all vector types


See merge request !36

Fixes #41

This allows to easily change the type of elements in a vector
expression, which can be useful to force a specific type when combining
vectors with different element types. For example, a double-vector can
be cast to float to avoid expensive computations in double precision.