Resolve inconsistency between dplyr and galah implementations of collapse() and compute()

[mjwestgate]

As of version 2.0, you can use collapse() to convert a data_request into a query_set. A query_set often contains >1 query, with the final query being the one requested by the user, and earlier querys being needed to check that it is valid. Finally, calling compute() evaluates the query_set into a single, valid URL:

x <- galah_call() |>
    filter(year == 2010) |>
    count() |>
    collapse()
x
Object of class `query_set` containing 3 queries:
• metadata/fields data: galah:::check_internal_cache()$fields
• metadata/assertions data: galah:::check_internal_cache()$assertions
• data/occurrences-count url: https://biocache-ws.ala.org.au/ws/occurrences/sea...

compute(x)$url
[1] "https://biocache-ws.ala.org.au/ws/occurrences/search?fq=%28year%3A%222010%22%29&disableAllQualityFilters=true&pageSize=0"

For occurrence queries, however, compute() is actually doing three things:

• evaluating (some) API calls to check the request is valid
• building a valid url
• sending that API call to the selected atlas

This is not ideal, because it means that for occurrence queries, it is impossible to interrogate the URL before sending, or to see it once it has been sent. Fundamentally, this problem occurs because of a conflict between the principle that collapse() shouldn't ping an API, and the need in galah to run checks before building a URL. I see two possible solutions:

• remove the limitation on collapse() to not hit any APIs; then collapse() would always return a valid query (or error if needed), but would also run checks
• add a function between collapse() and compute() (perhaps precompute()?) to build a single URL, but not push the request to the selected atlas

My instinct is that the former is neater, though it does change some behaviour from version 2.0. For context, this is what dplyr says about these functions:

compute() stores results in a remote temporary table. collect() retrieves data 
into a local tibble. collapse() is slightly different: it doesn't force computation, 
but instead forces generation of the SQL query. This is sometimes needed to 
work around bugs in dplyr's SQL generation.

[mjwestgate]

Decision on 17-01-2024:

• collapse() will now return a query object, which by default has slots type and url
  • i.e. previous behaviour of compute()
• collapse() also gains a new .expand (?) argument
  • when TRUE, this appends the query_set to an additional slot (possible called calls?) to support debugging
• compute() has reduced functionality under this model
  • mainly acts as a pass-trough function between collapse() and compute()
  • compute() is still critical for type = "occurrences", where it submits the query (but doesn't wait for a response)

[mjwestgate]

Implemented in v 2.0.1