squeeze for sparse matrices

[martinholters]

Before having SparseVectors, the default squeeze from abstractarray.jl returned wrong results (see #10813). Hence a dedicated squeeze implementation for sparse matrices throwing an error was added in f8e343e. Thanks to SparseVector, things look different now and that change can be reverted. (Adding functionality by removing code - fantastic!)

The only thing that doesn't work is squeezing down to 0 dimensions (e.g. squeeze(speye(1),(1,2))), if anyone needs that. It still errors, but with a less helpful message.

[tkelman]

Can you add a test? Thanks!

[martinholters]

@tkelman Good point, test is in.

[tkelman]

apt-get travis issue addressed by #15249, restarted

dpkg: dependency problems prevent configuration of gfortran-5:i386:
 gfortran-5:i386 depends on gcc-5 (= 5.2.1-23ubuntu1~12.04); however:
  Package gcc-5:i386 is not configured yet.
dpkg: error processing gfortran-5:i386 (--configure):
 dependency problems - leaving unconfigured
Setting up libdpkg-perl (1.16.1.2ubuntu7.7) ...
Setting up zlib1g-dev:i386 (1:1.2.3.4.dfsg-3ubuntu4) ...
Setting up libssl-dev:i386 (1.0.1-4ubuntu5.34) ...
Setting up make:i386 (3.81-8.1ubuntu1.1) ...
Processing triggers for libc-bin ...
ldconfig deferred processing now taking place
Errors were encountered while processing:
 gcc-5:i386
 g++-5:i386
 gfortran-5:i386
E: Sub-process /usr/bin/dpkg returned an error code (1)

edit: nevermind, it would require rebasing since the fix was in .travis.yml. Not worth the trouble, this passed on win32 so it should probably pass on linux 32 as well.