The -cpp option interferes with CMake's Ninja generator.

Also simplify literal types in the test code: this cleans up a number of
warnings about conversions, including an unintentional conversion to
single-precision reals (the literal 3.0 was used).

Also commented out unused functions that were generating warnings.

This also removes a couple of functions that were returning arrayviews
*by reference* (e.g. getGlobalRowView), so the wrappers that were in
place actually did nothing.

Gets rid of the 'FIXME' since SWIG now knows about Import when
instantiating Export.

Not clear why this was working in GCC. Whenever a SWIG execption function
fails, it returns an empty/null value, which in the case of a SwigArray
is `{null, 0}`. It's unspecified behavior to do `c_f_pointer` with
zero-sized arrays. GCC was somehow working, but flang chose to convert
this result to a null fortran pointer, which then is *assigned* to `iarr`
before the throw condition is checked.

The solution is to have the test return pointers rather than automatically
allocating arrays.

Updated to latest SWIG. Currently fails to compile:

```
       call A%getGlobalRowView(gblrow, cgview(1:nnz), csview(1:nnz))
                                                                   1
Error: Type mismatch in argument 'indices' at (1); passed INTEGER(8) to CLASS(swigtype_teuchos__arrayviewt_long_long_const_t)
```

Also clearing initial pointer assignments in the test.

- Changes intent of mutable classes to intent(in) since we only modify
the pointed-to-class, not the pointer
- Fixes modification of "other" pointer during assignment, which would
access possibly off-limits Fortran memory

- Updates memory management/assignment semantics
- Fixes some latent type issues