Travis: suppress logs for installing numpy, nose

[certik]

Previously one had to go through the logs and try to figure out where the log
for matplotlib starts. Since we are not testing that numpy builds here, it's ok
to simply use the "-q" option, that way the logs will not be cluttered by the
numpy build.

[dmcdougall]

That is so much better! I don't see any reason why we need the numpy logs. I also don't see any reason why we need all the "copying image_blah.png ..." messages either, but I think that will require more work. Ideally the best case scenario would be to have the compilation messages for the sources that need to be compiled, and the nose test output.

[mdboom]

As discussed in #1179, there are cases where seeing the Numpy logs is important, and we can't predict those cases. From experience, I've always been of the opinion that too much output (that is easy to filter) is always better than two little when it comes to CI logs.

One of the highly limiting things about Travis is the lack of a mechanism to store build products -- if it had that, I'd say these logs could go off into another file somewhere. But the real reason we need build products is that when image comparisons fail, and those failures aren't reproducible locally, debugging is extremely difficult. I have an aggregious hack in #1250 that tar/bz2/binhex encodes the failed images and puts them in the log so they can later be extracted. This was the only way I was able to determine that our earlier failures were due to 32-bit/64-bit differences.

All that said, on further consideration, it seems like the cases where Numpy will fail is rare, so I'm probably +0.5 on this.

[WeatherGod]

I am also with @mdboom on this, the install output is valuable. Perhaps as
a compromise, we should have the build output have easily grep-able lines
that would delimit the beginning and end of the different sections of the
logs?

[pwuertz]

Why not put them into the compressed archive Michael is building for the image diffs.

[certik]

Ah I see, I didn't realize that numpy failures can cause matplotlib failures. In my project, numpy is just one of the many prerequisites, so I am installing them with:

      pip install -q --use-mirrors cython numpy sympy;
      pip install -q --use-mirrors matplotlib;

And if there ever were failures due to numpy or cython or sympy or matplotlib, I can always setup a branch and delete the "-q" from the output. Most of the time (=all the time so far) I only want to see the logs from my project.

So I am closing this PR.