Don't use .save() in validate_version_metadata

[jjnesbitt]

We're seeing a very prevalent race condition, resulting in version asset summary aggregations being skipped (this can be observed via copious amounts of Skipped updating assetsSummary for version *** in the papertrail logs).

Here's what I believe is happening:

1. A version is updated in some way, and status set to PENDING
2. The scheduled task validate_draft_version_metadata is run, dispatching the validate_version_metadata and aggregate_assets_summary tasks at the same time
3. validate_version_metadata finishes before aggregate_assets_summary, calling Version.save(), which calls Version._populate_metadata, potentially modifying the metadata
4. aggregate_assets_summary finishes generating the summary and reaches this line, which will only update the version if the metadata field hasn't changed from when the task was dispatched. Since the above step modifies the metadata, this filter fails and the update is skipped
5. The version assets summary remains outdated/incorrect

This PR changes the behavior of validate_version_metadata to use .update when updating the status and validation_errors of a Version, subverting the underlying call to Version._populate_metadata, and thus any metadata changes.

@danlamanna I'd like to hear your input on this, particularly to make sure I'm not violating any intended behavior. I believe I've largely kept in line with what was already there, but I'm unfamiliar with the reason/need for using the metadata=version.metadata filter when updating a version's asset summary.

[jjnesbitt]

I just realized I may need to also include an update to the version modified field. This has been fixed.

[dandibot]

🚀 PR was released in v0.3.76 🚀