fix dimension issue in sampleDPmeasure

[paciorek]

sampleDPmeasure determines number of node elements in the mixture component variables based on the "data length" raised to the dimension of the variable.

This would work for something like y[1:3] with mu[1:3, z] and Sigma[1:3,1:3,z]. But not if mixing over regression components such as mu[1:3,i] <- X[1:7,i] %*% beta[1:7,z].

Not sure why we didn't originally just check the number of scalar components of each variable, but this PR does that.