Allow disabling DCP warnings

[ericphanson]

Fixes #362

I'm not a big fan of globals (and I spent some time getting rid of them in #322), but since we emit this warning even on things like x = Variable; y = Variable(); x*y, there doesn't seem like a good place to pass in an argument warn=false. Arguably we could only do the check when calculating the vexity of a Problem, and then pass warning info through the problem constructor. (We would need to also make sure you can see the check before starting a solve, etc..)

However, (1) the way the vexity of a Problem is written currently is a little strange and I'd prefer not to mess with it yet, and (2) it might be helpful that we are very eager to warn on non-DCPness (i.e., on x*y instead of waiting for the user to put that inside a Problem or call a solve!), because it can help formulate a DCP problem easily, checking along the way, especially when you are first learning the rules.

So for these reasons, I think a simple fix to the current way of emitting the warnings is best, namely to just check if warnings are enabled from a global before issuing one. In the interests of thread safety (since that appears to be a useful thing to do with Convex: #370), I followed the pattern of https://discourse.julialang.org/t/attempting-to-make-my-module-code-threadsafe/30434/2 of making a vector of parameters indexed by threadid(). I think the same pattern is also used for Julia's RNG. I used atomics, and added some for the other globals (for printing width and depth).