blocks_first_seen race metric may be inaccurate on CL→mump2p path

[swarna1101]

Context

trackedSlots in pkg/service/bootstrapper/beacon_block_measures.go is updated from two paths:

• composeBlockTelemetry — sets EthSeenAtMs / MumSeenAtMs (block seen)
• RecordMumPublishedAt — sets only MumPublishedAtMs (block published to mump2p on the CL path)

The blocks_first_seen_{mump2p,libp2p}_total counters (dashboard: "mump2p boost %") use map presence as "already seen":

firstForSlot := true
Upsert(slot, func(v) { firstForSlot = false; ... }, zeroVal)
if firstForSlot { IncBlocksFirstSeen(source) }

Problem

On the CL path, RecordMumPublishedAt runs synchronously after publish, while HandleBeaconBlock → composeBlockTelemetry is async. The map entry can exist (with both seen fields still 0) before the libp2p seen timestamp is written. LibP2P may then miss first-seen credit even when it arrived first → possible undercount of libp2p-first / inflated mump2p boost %.

Open question: should an entry ever exist with both EthSeenAtMs == 0 and MumSeenAtMs == 0? Today it can, because RecordMumPublishedAt creates one.

Options

1. Count first-seen when EthSeenAtMs == 0 && MumSeenAtMs == 0
2. Store publish times separately so trackedSlots presence means seen
3. Leave as-is if the race is negligible in practice

Note

An in-flight fix was reverted until we agree on intended semantics and validate dashboard impact.