We actually need a mix of cmake magic and config.h to work
* if alberta is not used (then the config.h define is taken)
* alberta is actually used (then the dimension has to be specified for the
  target at hand with add_dune_alberta_flags(GRIDDIM $dim $target))

This reverts commit 35aa624d.

This reverts commit db210150.

refactor la backends discovery/usage

Since we are pulling dune-fem, and thus dune-grid and dune-alugrid in turn,
the unspecialized DGF parser is pulled, which does not compile with
dune-alugrid.

Unfortunately, alberta headers globally defines TEST, which leads to tons
of compiler warnings. Te clean solution is to not let gtest define TEST.
This unfortunately requires us to write GTEST_TEST instead of TEST.

# Conflicts:
#	.gitignore
#	.travis.yml
#	dune.module

This did not compile for me with clang 3.8.1-12~bpo8+1 using gcc
4.9.2-10 stdlib. Any objections against this change, @dune-xt-common-devs?
Was there any reason to use int instead of double in the first place?

Previously, TBB_INCLUDE_DIR was ignored, although documented as possible
hint to where the tbb include might be. No it is respected, hopefully
without breaking findidn of system installation ...

The old behavior only added `-lboost_system` but did not add
`-L/path/to/boost/lib`.

Previously, TBB_INCLUDE_DIR was ignored, although documented as possible
hint to where the tbb include might be. No it is respected, hopefully
without breaking findidn of system installation ...

The old behavior only added `-lboost_system` but did not add
`-L/path/to/boost/lib`.