Extend test compilation support to include Algorithm layer

[arporter]

Currently the support for compiling code generated during tests is limited to the PSy layer, or standalone pieces of code (i.e. with no dependencies). We have no way of testing that the manipulated Algorithm code compiles.

Most of the Algorithm tests call the generate() method which just returns fparser1 ASTs for the Algorithm and PSy layers. Compiling this code requires that we first compile the kernel routines and there's no easy way to identify what they are at this level.

[LonelyCat124]

We should also extent test compilation support to handle tests which create separate files, e.g. test_mod used for testing code that imports symbols or similar.

[arporter]

We could potentially use https://github.com/stfc/fparser/blob/master/example/create_dependencies.py to generate a Makefile to build the files in the tmp directory.

[arporter]

Or, we just let the caller specify a list of filenames in the order in which they must be compiled?

[arporter]

@mo-alistairp and I have just been burnt by this. It turns out the implementation of ScalarArray arguments wasn't complete but this wasn't picked up by the tests. (It would if we had added an example but we omitted to do that.)

[arporter]

Now that we have the ~full LFRic infrastructure, we should be able to do this. Currently the utilities.py::Compile._code_compiles only takes a PSy object. Possibly we want an alg_compiles although that's going to get very close to just "run PSyclone, build the generated files"?