fix(nimbus): rename event_stream to events_stream, add SourceType enum, add ratios to FeatureSpec

[yashikakhurana]

Because

• events_stream matches the actual BigQuery table name; event_stream was inconsistent
• ratios was referenced in the bigquery-etl query template but had no way to be defined in TOML configs
• The source type was validated against a plain frozenset with no type safety

This commit

• Renames event_stream → events_stream in VALID_SOURCE_TYPES, SourceTableSpec default, and firefox_desktop.toml
• Adds SourceType enum (metrics, events_stream, clients_daily) for type-safe validation
• Adds ratios: list[list[str]] to FeatureSpec so TOML configs can define metric ratios
• Updates FeatmonSpec.from_dict to populate ratios from config
• Bumps version to 2026.4.4

Related: EXP-6732

[github-actions]

⚠️ Changing default metrics will cause all experiments that are currently live to get rerun after this change has been merged. This may come with a substantial cost. To skip reruns, add [ci skip-rerun] to the PR message.

[github-actions]

✅ Jetstream Validation is complete. Check the CI logs for this step for Query SQL and data processing estimates.

[github-actions]

⚠️ Experiment metrics query data processing estimate exceeded high threshold and may timeout during analysis.

This does not block merging, but experiment analysis may timeout and require config changes.

[yashikakhurana]

@mikewilli The integration tests clone firefox_desktop.toml from the main branch on GitHub rather than from this PR branch, so they were loading type = "event_stream" while the updated validator now only accepts events_stream — causing the failures.

To unblock CI, I've added event_stream as a deprecated alias in SourceType. Once this PR merges, main will have type = "events_stream" in the TOML and the alias can be removed in a follow-up. Both the parser change and the TOML change are in this PR, so they land together.

[mikewilli]

Made a couple comments, but everything here looks good.

Did you have something changed in the jetstream/ dir at some point? I'm wondering why the CI put in some messages about experiment analysis, that should only run when jetstream/** stuff is changed.

Also, the integration test issue seems like a bad thing that we should fix (separately), I'll have to think about it. If you're happy with landing this and then updating again afterwards to remove the deprecated event_stream option, then I don't think we need to block this. Thanks!

[mikewilli]

You might consider waiting on this. You can have an immediately follow-up to remove the deprecated event_stream option that then pushes a new version.

[yashikakhurana]

Made a couple comments, but everything here looks good.

Did you have something changed in the jetstream/ dir at some point? I'm wondering why the CI put in some messages about experiment analysis, that should only run when jetstream/** stuff is changed.

Also, the integration test issue seems like a bad thing that we should fix (separately), I'll have to think about it. If you're happy with landing this and then updating again afterwards to remove the deprecated event_stream option, then I don't think we need to block this. Thanks!

yeah my bad did added some changes which were not required