Speed improvements

[gusbeane]

Some speed improvements to brutus. Might require slight tweaks to conform to your linter.

Major changes:

• Replace scipy's truncnorm.logpdf and chi2.logpdf with a self-written version
• Replace multivariate sampler with a self-written version

Minor changes:

• Switch to Fortran ordering for some arrays
• Use np.random_intel when available (this could be removed, when I tested with the intel distribution it did not make any difference in speed - however, there may be further changes that need to be made to the code to make it really work, so I would advocate keeping it for now and make a decision at version upgrade)

I recommend not to upgrade to v0.8.2 until we have discussed further possible changes to _optimize_fit

[joshspeagle]

Okay, I'm going to pull this and then do some small cleanup.