not rescaling stdevfacI and stdevfacC to mean 1 should cause a warning

[casperwberg]

When stdevfac's are used without rescaling to mean 1 implies
priors on variance parameters means something different,
for example if stdevfacI is a vector of 2's then the estimate of sdi is halved.

automatic rescaling would break backward compatibility, so giving a warning is probably best solution.