paper_trail is inconsistent when Version records cannot be created. in some cases (create) an error is raised, in others (update, update_columns, and destroy) an error is logged but not raised. this can result in history silently being lost which is certainly unexpected behavior.

Currently when creating a new versioned record an error is raised if the Version cannot be created - this is because #save! is used:

However, when updating, using update_columns, or destroying a versioned record no error is raised if the Version cannot be created:

proposal

i think the behavior should be consistent - all failures should either raise errors or not. personally i don't see any advantage to not raising an error in these cases - it likely means that whodunnit isn't defined which i would expect to be an oversight rather than intentional. i realize this is not backwards compatible so i propose a two step process:

1. an opt-in configuration parameter is added to the current major version which causes update and destroy failures to raise an error.
2. in the next major version bump this becomes the default behavior.

does this seem reasonable? i'm happy to take this on asap.