feat(eap): Implement sequence number offsets for eap items

[Dav1dde]

Implements INGEST-804

[linear-code]

INGEST-804

[jjbayer]

Suggested change

	/// This must only run once, to not shift the same timestamp multiple times and therefor should
	/// be limited to processing Relays.
	/// This must only run once, to not shift the same timestamp multiple times and therefore should
	/// be limited to processing Relays.

Would it make sense to defensively only modify the timestamp if its sub-millisecond part is zero? Because there's no need to apply it otherwise, right? Or alternatively check if there's already a remark for timestamp.sequence.

[Dav1dde]

Would it make sense to defensively only modify the timestamp if its sub-millisecond part is zero?

This won't work if an SDK sends the timestamp as rfc3339 which we parse correctly with ns precision.

Or alternatively check if there's already a remark for timestamp.sequence.

That's a possibility, but then we would apply the shift differently depending on the amount of Relay instances in the chain, due to clock drift correction running in every Relay.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 3bbd9d1. Configure here.}

[loewenheim]

Could we potentially resolve this for spans (if it ever becomes necessary, not now) by having reference_timestamp(_mut) return an iterator of timestamps?

[Dav1dde]

Something like that was my idea, I was though thinking instead of applying it separately, there is maybe a way where we can adjust the clockdrift processor or a separate processor which applies the offset to all timestamps equally.