Resolve validation step info to string for hashing

[yjunechoe]

Summary

This PR introduces a new internal function hash_validation_step() which ensures that functions (in preconditions) and quosures (vars() in value/left/right) are resolved to character for the purposes of hashing. In turn, digest::sha1() is guaranteed to always receive a character vector - this resolves issues with environment hashing as described in #509.

Currently, the new hashing implementation replaces the previous one (vs. being opt-in).

Pending

• Backwards compatibility for older agents in the multiagent: we would need to re-hash agents' steps for the purposes of alignment.

• More tests to check that this new method does not create false positives in step alignments. Alignment should respect identity up to yaml-equivalence.

[yjunechoe]

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 Since updates to dplyr can make snaps outdated (therefore causing tests to fail unexpectedly), I've commented that out in the draft_validation tests (so the PR will also resolve the testthat GHA failing on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).

[yjunechoe]

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.

[rich-iannone]

LGTM!

[rich-iannone]

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