allow no-k model for multiple searches

[juniperlsimonis]

@ddalthorp 's suggestion from #136
Error in pkm(p ~ Visibility * Season, data = pkdat[pkdat$Size == "small", : No formula or fixed value provided for k. Provide one. ...Change to warning that no k is provided and tag the result to indicate "no K" so not allowed in estimation of g. A use for a "no k" model might be to look at how p changes with search occasion but not constrain it to a k-model. This would be a research or curiosity question rather than an M-estimation question, though, and we should require an estimate or decision about k before fitting models for g.

[juniperlsimonis]

does the mechanics of the estimation procedure change in this situation, @ddalthorp ?
like what needs to happen to make the model fit p on each search occasion? (if it's simple, i can include it in the current edits i'm doing)

[ddalthorp]

Fitting p on one search occasion (or several different search occasions in succession) with no k is simple.

The current modeling approach handles this just fine if done iteratively by search occasion, but we get a series of estimates rather than one, i.e., for the first search, phat[1] = pkm(just using data from search 1); and phat[2] = pkm(just using data from search 2). This should work just fine in most cases (but collapses when y = 0 or y = n).