Test for error output

[saper]

Enable testing for error output
and sass compiler failures.

If files "error" and "status" are existing
in the test directory, do expect errors
and test against them.

Errors returned by the Sass engines are
now treated as test failures, not errors.

Also create error and status files with --nuke.
An error will be reported in this case anyway.

Important Now the tests cannot be ran in parallel,
because we need to capture standard error reliably.

Should fix: #136 #133

[saper]

As the Travis shows, this change work only with adding compiler via the -c option (can be sassc or Ruby sass CLI). Unfortunately, calling Ruby sass internally presents the incompatible interface (this needs to be fixed, too).

[saper]

So after resolving #495, #497, #499 the tests finally pass...

[saper]

This should be enough for rudimentary error testing. Should we need more, we should be aware of the following limitations:

• status returned from external process is not numeric ( EngineAdapters: inconsistent status value returned #495 ) - solved via ExecutableEngineAdapter should return numeric status #496 (merged)
• status returned from sass cli, sassc, and Sass ruby module have no consistent values ( EngineAdapters: inconsistent status value returned #495 ) - worked around via saper@6495e47 by checking only for 0 (success) and non-zero (error)
• no consistent form to report exceptions between engines ( EngineAdapters inconsistently report exceptions #497 ) - worked around by picking the first line only saper@66bdbdd
• don't send CSS-formatted exception string to stdout - Return an exception message to stderr #498 (merged)
• race condition trying to capture Sass ruby module std err output ( Error output: SassEngineAdapter cannot be reliably parallelized #499 ) - worked around by turning off parallel testing saper@2239245