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.

Split ExpressionL1Norm, ExpressionL2Norm from ExpressionLpNorm because they have different result type

Also fixed the result types and starting values for other vertical operations.

It is very slow and unnecessary - the user can easily use operator[] or
getElement and setElement for any operation with a single element, not
just addition.

Distributed argMin and argMax are not implemented yet, all other
operations successfully pass all tests (which were unified with the
tests for Vector, VectorView and StaticVector).

Simplified expression templates: pow implemented as a binary expression rather than a special unary expression

Does not work yet for the distributed vector...

getConstView should be always used for const operations, because
explicit is better than implicit. Overloading one method for cases which
have a completely different type (ConstViewType is *not* just a
const-qualified ViewType) can be very unclear (or even ambiguous) in
certain cases.

This way we can build with CUDA, Clang and MPI at the same time

unsigned int cannot hold a value of 0.5, it was probably meant to wait
0.5 seconds, i.e. 0.5e6 microseconds.

[ci skip]