Introduce construct.model parameter

[luciferous]

construct.model is a function that takes modeling data (NA in
post.period) and returns a bsts model.

This invocation:

go <- function(data) {
  y <- data[, 1]
  sdy <- sd(y, na.rm = TRUE)
  sd.prior <- SdPrior(sigma.guess = 0.01 * sdy,
                      upper.limit = sdy,
                      sample.size = 32)
  ss <- AddLocalLevel(list(), y, sigma.prior = sd.prior)
  bsts.model <- bsts(y, state.specification = ss, niter = 100,
                     seed = 1, ping = 0)
}

impact <- CausalImpact(data, pre.period, post.period,
                       construct.model = go)

is equivalent to:

model.args <- list(niter = 100)
impact <- CausalImpact(data, pre.period, post.period, model.args)

. This change provides an interface that has RunWithData's flexibility
with pre- and post-period, and RunWithBstsModel's full configurability
of the model. For example, the caller can now specify a custom model
where the post-period is before the end of the data, or with a gap
between the pre-period and post-period, either of which aren't possible
with the previous interface.

[luciferous]

plot(impact) from the added test case.

[luciferous]

Hi @alhauser,

This proposal changes the API but in a backwards-compatible way. Let me know if you'd be open to it, I'm happy to chat about other ways to get the same outcome.