chore: Add initialization tracking for FDv2 Data Source.

[kinyoklion]

Note

Introduces initialization tracking for FDv2 with observer-driven composites, wraps the data source to complete on initialization, and updates polling/unrecoverable error behavior and tests.

• FDv2 core:
  • Add InitializationTracker, InitializationObserver, and CompositeObserver to aggregate signals from initializers/synchronizers and determine init completion.
  • Return a new wrapper CompletingDataSource from FDv2DataSource.CreateFDv2DataSource that completes Start() based on aggregated initialization rather than inner source start.
  • Refactor FDv2DataSource into Internal/FDv2DataSources namespace; replace direct applier wiring with observer composition; add DataSourceCategory classification.
• Composite/Composition:
  • Integrate observers into CompositeSource usage; propagate status through ObservableDataSourceUpdates; continue to shut down with Off and exhaustion error when sources are depleted.
• Polling data source:
  • On unrecoverable HTTP errors (e.g., 401), set init task result to false and shut down; continue to mark Valid on 304 responses.
• Data system:
  • Wire FDv2DataSystem to new Internal.FDv2DataSources types.
• Tests/contract:
  • Update unit tests to expect Start() result false on unrecoverable errors and true when no sources; adjust expectations for initialization/fallback flows.
  • Contract test suppressions trimmed; keep streaming/fdv2/disconnects on goodbye.

^{Written by Cursor Bugbot for commit bba9848. This will update automatically on new commits. Configure here.}

[tanderson-ld]

Is this reason this has to be tracked separately from Synchronizers is because the Fallback composite has its own FactoryList that exhausts separately from synchronizers?

[kinyoklion]

Yeah, without it we would end up basically in an orphaned state, or exit too early. Because if it is just a synchronizers, then the synchronizers would be exhausted twice.

The other thing is that each state is contingent on available states. Like if we don't fallback, we cannot wait for fallback to be exhausted. Or if we don't have synchronizers, etc.

[cursor]

Bug: Race condition in FDv1 fallback initialization tracking

When synchronizers fail with the FDv1Fallback flag, the CompositeObserver calls FDv1FallbackActionApplier first (which starts fdv1Synchronizers asynchronously), then calls synchronizationObserver second (which immediately sends SynchronizersExhausted). Since FallingBack is only sent when fdv1Synchronizers report Initializing state (asynchronously), SynchronizersExhausted is processed first. In InitializersExhausted state, receiving SynchronizersExhausted calls HandleRemainingSources() which sets state to Failed if both _initializersRemain and _synchronizersRemain are false. The comment at line 122 says "We only consider fallback if we have transitioned to fallback" but the implementation doesn't check the current state, causing premature failure before fdv1Synchronizers can provide data.

Additional Locations (1)

• pkgs/sdk/server/src/Internal/FDv2DataSources/FDv2DataSource.cs#L95-L97