Re-use error covariance structures from STATO

[cmaumet]

Following discussions at ISA-tools/stato#28, this is a proposal to:

• re-use stato: 'covariance structure' in place of nidm:ErrorDependence.
• re-use stato: 'compound symmetry covariance structure' in place of nidm:CompoundSymmetricNoise.

@nicholst: could you help me in finding out if there is a match for some of these nidm terms:

in the current set of covariance structures defined in STATO:

(Once the discussion will be fully implemented at ISA-tools/stato#28, we will replace all nidm terms currently defined as children of stato:'covariance structure'.)

[nicholst]

@cmaumet, surprisingly, these don't map so well.

• nidm:ArbitrarilyCorrelatedError: amazingly, STATO seems to have omitted SAS's "Unstructured" covariance structure. I've raised this with an addition to STATO's Dependence structure ISA-tools/stato#28. So, no reference for now.
• nidm:ExchangeableError: This can reference STATO's "compound symmetry", but in fact is more general -- compound symmetry is just about correlations/covariances, while "exchangeability" refers to the entire multivariate distribution. "Gaussian variables that are exchangeable are compound symmetric."
• nidm:IndependentError - Again, no love in STATO; raised in my Dependence structure ISA-tools/stato#28 update.
• nidm:SeriallyCorrelatedError - Finally! This maps exactly to 'Toeplitz covariance structure'

[jbpoline]

I was a little puzzled as well.

+1 For SeriallyCorrelatedError == Toeplitz. (in case of stationarity, always assumed for fMRI...)

[cmaumet]

Thank you @nicholst and @jbpoline! Re-use of stato:'Toeplitz covariance structure' is now implemented. #312 is open to remind us to replace the remaining error dependence terms once ISA-tools/stato#28 is closed.

I will now merge those changes.