which only compiles with gcc atm

New complex<double> tests still fail though.

these have a lot better output on failure than
EXPECT_TRUE(FloatCmp::XX(a,b))

simply uses the same distribution, generator and limits for real and
imaginary parts

and forwards to each Visitor's run(...) accordingly

now with less code, fewer exception throwing and more test coverage ;-)

closes #63

thanks to @tobiasleibner for spotting the problem

This is again in line with treatin complex as a vector and lets the
float compare tests pass again.

Complex numbers can take the form `+-1.1+-2.2i`, `+-1.1`, but not
`+-1.1i`. That last one would need to be written as `0+-1.1i`.

There was an unannounced interface change in Parametertree which this
commit works around, while keeping backward compat.

offender was in gtest

also harmonizes from_string arg passing

test creation failed to provide values to compare against. This is now
automated via the VectorAbstraction. VectorAbstraction::create(size=s,
val=y) will create a complex number with real(complex) == y ==
imag(complex). Imo that's consistent with the rest, esp since we decided
to treat complex numbers as vectors anyhow.

Linear function

Add threading.max_count to global config when setting it

refs #59 and also tests float_cmp for actual floats

refs fa6bc47665039871bd256dbba230c1bbae014077

These are actually copy and movable, but not assignable. Note that this
holds for the providers here, not the actually wrapped storage!

The same assert is present in the base above.

This removes the need to include fvector.hh