Switch CaKernel code from template gpu_cuda/3dblock to gpu_cuda_dc/3dblock2.

[dmkoppelman]

In this issue have Kranc generate non-boundary CaKernel code using
template gpu_cuda_dc/3dblock2 (it currently uses gpu_cuda/3dblock) and
make associated changes. This template is the template that is
actively being developed it performs dynamic compilation by default
(which increases startup time) but that can be turned off using the
parameter:

CaCUDALib::dynamic_compilation = no

This issue will also fix two problems that have occurred in
SimpleWave. The CaKernel.ccl file can have a wrong value for STENCIL
and EXTERIOR. The STENCIL can be wrong because CalculationStencilSize
is used on calculations which are not in the right form (the
differencing operators have not been sufficiently expanded). The
EXTERIOR is set based on the Where value, which is a Kranc input. A
possible value is Automatic, which the code does not try to
resolve. Both of those problems will be fixed by using the values for
Where and stencil size that are computed in
CreateCalculationFunction. The reference functionality added in Issue
113 will be used to move those values up to KrancThorn.m without
extensive refactoring.

So far, these changes fix problems with SimpleWave. That is,
it appears to generate correct results. I've ran Kranc on ML_BSSN,
but I haven't tried building it on Cactus.