Proposal

Part of #14217

With the PRW 2.0 spec comes implementation and metadata per series logic. This unblocks using it to track CT without adding zero sample which has some drawbacks (e.g. potentially misleading synthetic sample that was NOT added by scrape and inability to use this in remote write).

Proposed here.

Initially we thought we could reuse metadata records, but another idea with sample record change is also viable.

Acceptance Criteria

• CT is being tracked in WAL, clearly distinguished from zeros.
• Behind feature flag (different than created-timestamp-zero-ingestion as those features might be considered orthogonal, or even colliding).