Add more loglikelihoods (ARMA variants and others)

[ben18785]

Additive ARMA(1, 1) errors are just one of a slew of potential error models to consider including. Others include:

• Heteroscedastic errors where errors scale with value of y
• Multiplicative independent and ARMA errors
• ARMA processes using Student-t errors
• Higher order ARMA(p, q) processes
• Autoregressive-conditional-heteroscedacity (ARCH) and GARCH (G for general) processes, which allow an error process which changes with time
• Kalman filter state-space models (probably a whole student project there...)

In principle, #661 should highlight which, if any, of these processes is appropriate for data.

[MichaelClerx]

@ben18785 @rcw5890 is this still something we want to do?