Better alignment of agent$validation_step to rows of get_agent_report(agent)

[yjunechoe]

This is a small surgery on a single line of get_agent_report():

pointblank/R/get_agent_report.R

Line 451 in 5a00f4b

validation_set <- validation_set[report_tbl$i, ]

The PR just makes the logic of matching rows between validation_set and report_tbl more explicit, via filter() on the matching i columns (where agents always have the invariant of all(diff(agent$validation_set$i) == 1)).

This resolves a cryptic error you get when you remove rows from the agent$validation_set for the purposes of get_agent_report() (ref: #563). Now, we get a better alignment of "rows in $validation_set" and "rows that appear in the {gt} agent report". I'm taking caution to not advertise this as public API (IMO getting the "hide rows" features in {gt} will serve as a better interface to this, when that's implemented). For the time being, the PR simply addresses some confusion about why removing rows in $validation_set does not produce the same effect for get_agent_report() (with an added benefit of making the intent behind the code for this line more explicit).

Using reprex from #563 (hide inactive columns), this now works:

devtools::load_all()

x <- iris |> 
  create_agent() |> 
  col_exists("Petal.Length",
             active = has_columns(iris, Petal.Length)) |> 
  col_exists("Spec",
             active = has_columns(iris, Spec)) |> 
  col_exists("Sepal.Length",
             active = has_columns(iris, Sepal.Length)) |> 
  interrogate()

get_agent_report(x)

x$validation_set <- x$validation_set[x$validation_set$eval_active, ]
get_agent_report(x)

[rich-iannone]

LGTM!