FIX: update.brmsfit returns error when updating an empty fit

[trinhdhk]

An empty fit is useful when users make post-hoc edit on Stancode but still want to leverage brmsmethods like standata.
Currently update.brmsfit throws S4 error when trying to access object$fit@sim. I think this is R's design inconsistency when $ is a safe call but @ isn't.

This pull request fixes this issue by implementing a check on object$fit.

empty_fit <- !isS4(object$fit)

Test:

m0 <- brm(mpg~cyl, data=mtcars, empty=TRUE)
mtcars2 <- dplyr::mutate(mtcars, cyl=1)
m1 <- update(m0, newdata=mtcars2, empty=TRUE)
m1; head(m1$data)

Expected output:

• m1 should be printed properly as an empty fit
• m1$data$cyl should be 1

 Family: gaussian 
  Links: mu = identity 
Formula: mpg ~ cyl 
   Data: mtcars2 (Number of observations: 32) 

The model does not contain posterior draws.
                   mpg cyl
Mazda RX4         21.0   1
Mazda RX4 Wag     21.0   1
Datsun 710        22.8   1
Hornet 4 Drive    21.4   1
Hornet Sportabout 18.7   1
Valiant           18.1   1

Test for recomplication:

m1 <- update(m0, newdata=mtcars2)

Expected output:

• The message "The desired updates require recompiling the model" should not appear.

Compiling Stan program...

---

FIXME:

I also put some safeguard in update.brmfit_mutiple but I did not use this function extensively enough to assure that is sufficient. From the surface I think it would err out when calling brm_multiple(empty=TRUE):

Compiling the C++ model
Fitting imputed model 1
Warning: Caught simpleError. Canceling all iterations ...
Error:
! unable to find an inherited method for function ‘get_stanmodel’ for signature ‘object = "NULL"’

I think it is not even allowed. Given the ancient cat "Compiling the C++ model" I believe it has been long since this function is updated.

[paul-buerkner]

Thanks! I will review this PR soon.