Add DIC about chainconsumer

[robert-crittenden]

2007 MNRAS paper by Andrew Liddle: Mon. Not. R. Astron. Soc. 377, L74–L78 (2007)

defines all these information criteria

[robert-crittenden]

Divergence information criterion (DIC):

Must evaluate:
bar{theta}, which is the expectation of the parameters over the posterior.
chisq (bar{theta}), the value of chi-squared at this mean parameter.
bar(chisq (theta)), the average chi-squared (expectation value) over the posterior

If I understand right,
DIC: 2* bar(chisq(theta)) - chisq (bar{theta}).
up to a constant that seems to depend on the data, so would not affect comparisons between models.

[Samreay]

Implemented in ChainConsumer v0.15.1