Validate user supplied TARGET

[TiborGY]

This has a tiny side effect, the error gets triggered even during make clean, but otherwise it works as intended.

[TiborGY]

About the Travis failures:
-The ARM builds have timed out, no idea why, maybe the Travis servers are too busy
-There is one legitimate failure. I do not really understand how the Travis builds work, but the build parameters seem to be conflicting. First there is

export BTYPE="BINARY=64 NO_AFFINITY=1 USE_OPENMP=0 NO_LAPACK=0 TARGET=core2"

After that,

$ COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"

Is this supposed to to be a CORE2 or a NEHALEM build? Also, why set DYNAMIC_ARCH=1, if you want a specific target?

[martin-frbg]

ARM tests are always timing out as the CI job runs on virtual hardware (qemu) - the idea is to use them for now to catch any glaring compile bugs, and set up automatic integration with a dedicated system later. (#1867).
The failing build is a good demonstration, setting TARGET in addition to DYNAMIC_ARCH can be useful when the build host is not autodetected, or to ensure that the common parts of the code (BLAS driver routines and the entire LAPACK) do not rely on "modern" cpu features present in the build system.