Fix warnings related to conversion to string literals to char*

Re-factor the linear solver and sparse integrator

Make solver interfaces more generic makes it easier to plug external linear solver in to the time stepping method.

Conflicts:
	src/util/interactiveDeformableSimulator/interactiveDeformableSimulator.cpp

Some changes to get things building on OS X again.

... when threads and OpenGL support are off.

It looks for the Xmu library even though it is
not a dependency of freeglut3.

... to see if it builds without BLAS/LAPACK.

... now that they are not duplicated in the `include` directory.

This also adds a configuration option named
`VegaFEM_ENABLE_OpenGL_SUPPORT` to match Ricardo's build.

Now you can turn on `VEGA_ENABLE_CG_SUPPORT` and
expect the build to complete.

Rename `BUILD_MODEL_REDUCTION` to `VEGA_BUILD_MODEL_REDUCTION`.
Add some include directories for "private" headers to get picked
up by dependent libraries; for example, the quaternion library
now marks its own source directory (`src/libquaternion`) as an
include directory for targets that depend on it. Note that this
is only for the build tree, not the installed version of Vega.

There is still some fiddliness going on with LAPACK; at times
CMake does not seem to want to link to it even though it is listed
as a dependency. The workaround seems to be upgrading to CMake 3.1.

Build static libraries on Windows until we start exporting symbols.
Keep the list of Vega modules in the top-level CMakeLists file and
simplify all the `add_subdirectory` calls.

VegaFEM now installs libraries plus CMake config
and target files so that it can be used by other packages.

PS. The duplicate headers in include and src/libXXX are evil!

PPS. The Vega getopts library has been renamed libvega-getopts
     because of issues with the preprocessor and potential
     install-name conflicts/confusion with the real getopt.