TNL is not supposed to have any external dependencies, so an empty
config file is useless. If some marginal parts of the project (such as
code for image processing) need to depend on some non-standard library,
they should be split into a separate module/project so that the
dependencies can be properly specified.

For now, the macros are specified as compiler parameters. Client
software using the affected headers will have to do the same.

fixes #13

fixes #18

fixes #23

fixes #21

Calculation of bandwidth broke in commit fb97bcd0

Check of MpiCommunicator::IsInitialized is necessary because
non-distributed benchmarks like tnl-benchmark-blas do not initialize
the communicator.

The former has been formally deprecated for a long time and in CUDA 10.0
it is finally properly marked as deprecated, so the compiler can print
warnings.

According to benchmarks, there is practically no difference in
performance. Only explicit unrolling is helpful, but that has been
implemented for the general algorithm in Reduction::reduce as well.

Now that error handling is done via exceptions, the return value is free
for more sensible things.