Handle source code FMU's better

[Martin Sjölund]

There are several issues that should be fixed w.r.t. FMU's:

• translateModelFMU corresponds to the behaviour of buildModel, and should be called buildModelFMU. translateModelFMU should create a source-code FMU.

We should always create a source-code FMU first, and then update the files within it to include the binaries for one target. So something like: buildModelFMU(M, version="2.0", targets={"linux32","linux64"}) actually corresponds to translateModelFMU(M, version="2.0"); fmuBuildBinary("M.fmu", "linux32"); fmuBuildBinary("M.fmu", "linux64"); (and if the target cannot cross-compile for one platform, you can send the FMU to that platform and compile the sources there instead).

When compiling a source-code FMU, two primary options are needed: linking the runtime, or compiling it yourself (necessary for targets that OpenModelica does not support; embedded targets, etc).

We also need to know which C sources are necessary in order to compile a source-code FMU (which is different for ME and CS).