Fix bonded tests

Fix bonded tests

These were not robust enough to pass everywhere (e.g. release
builds). They probably still are not, but fixing that needs more work
than we can do now. A key part of this is that acos implementations
differ, leading to different bond angles being computed, which leads
to different rounding error in accumulating forces and shift forces.
The latter now use different tolerances, because often we compute a shift
force that is zero, modulo rounding error.

Updated required version for reference build type to gcc 7.

Used namespaces better in bonded tests

Added some GoogleTest pretty-printers to see better what was going on.

Added useful strings describing bonded kernel flavors so test failures
had more context.

Used coordinates for butane in tests so that numerical instability was
less of a problem.

Avoided an unnecessary include of vectypes.h and testasserts.h