Also dplyr::storms turns out to be one of those data that get regularly updated. Since updates to dplyr can make snaps outdated (therefore causing tests to fail unexpectedly), I've commented that out in the draft_validation tests (so the PR will also resolve the testthat GHA failing on main).

I think this should do the trick!

Some notes on the implementation:

• Hashing logic is factored out into hash_validation_step(), which resolves everything to string before digest::sha1() is called. Current implementation appends "-v0.12" to the hash.
• An agent's steps can be re-hashed via another new internal function rehash_agent(), which iterates the above function over the rows of the agent's $validation_set. This returns an identical agent that only changes in $validation_set$sha1. It can be used in isolation to repair or update an agent's hash for validation steps:

agent1 <- create_agent(small_table) %>%
  rows_distinct() %>% 
  interrogate()
agent1$validation_set$sha1
#> [1] "adac947b3a26f7df4bfd374b5fff4e67b4190f8b-v0.12"
agent1$validation_set$sha1 <- "bad hash"
agent2 <- rehash_agent(agent1)
agent2$validation_set$sha1
#> [1] "adac947b3a26f7df4bfd374b5fff4e67b4190f8b-v0.12"

• create_multiagent() now rehashes all agents it receives, if it detects any bad or conflicting hash versions. Currently, it's either that all hash are v0.12 (doesn't trigger rehashing) or not (triggers rehash). We could expose an argument to let the user control the rehashing method, but maybe not necessary to introduce that at this stage (we only have 2 options and one is clearly better).

agent1$validation_set$sha1
#> [1] "bad hash"
agent2$validation_set$sha1
#> [1] "adac947b3a26f7df4bfd374b5fff4e67b4190f8b-v0.12"
multiagent <- create_multiagent(agent1, agent2)
multiagent$agents[[1]]$validation_set$sha1
#> [1] "adac947b3a26f7df4bfd374b5fff4e67b4190f8b-v0.12"
multiagent$agents[[2]]$validation_set$sha1
#> [1] "adac947b3a26f7df4bfd374b5fff4e67b4190f8b-v0.12"

get_multiagent_report(multiagent, display_mode = "wide")

All in all, this fixes both the alignment issue (#509) and the performance issue (#508). There are no user visible changes beyond "bugfix in multiagent alignment", so the news item for this PR just reports that.

This is really great work! Feel free to merge any time.