add dp_mixgauss cluster,

[xinglong-li]

add dp_mixgauss cluster (as well as the clustering using finite mixture);
revised the dp_mixgauss sample code to make it more readable

[coveralls]

Pull Request Test Coverage Report for Build 2581925963

• 0 of 137 (0.0%) changed or added relevant lines in 1 file are covered.
• 3 unchanged lines in 1 file lost coverage.
• Overall coverage decreased (-0.3%) to 11.038%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
probml_utils/dp_mixgauss_utils.py	0	137	0.0%

Files with Coverage Reduction	New Missed Lines	%
probml_utils/dp_mixgauss_utils.py	3	0%

Totals
Change from base Build 2566994900:	-0.3%
Covered Lines:	505
Relevant Lines:	4575

---

💛 - Coveralls

[murphyk]

This looks good. I will merge for now, but I wonder if we could create a unit test of some form? Perhaps if the true dataset contained only one cluster, we would expect the DP samples to eventually converge on the K=1 and the true mean and covariance?