metrics.md: add lean_aggregator_skipped_total + update aggregated_signatures_building_time status

[ch4r10t33r]

Summary

Adds a new cross-client metric lean_aggregator_skipped_total{reason=...} so operators can answer "how many slots did aggregation actually run for, and how many were skipped and why?" with a single counter rather than deriving the answer from coverage gauges or grepping logs.

Also updates the existing lean_pq_sig_aggregated_signatures_building_time_seconds row to reflect what's actually exposed on the live devnet (Grandine and Zeam).

Motivation

Today no client exposes a standard skip counter. Operators surveying the live devnet found:

• zeam exposes zeam_aggregate_skip_total{reason=not_aggregator|not_synced|missing_state|spawn_failed} (client-namespaced)
• ream / grandine / ethlambda / lantern expose nothing equivalent

That leaves "did this aggregator drop a slot?" as a derived question. The two indirect proxies — per-slot subnet coverage (lean_attestation_aggregate_coverage_subnets) and lean_pq_sig_aggregated_signatures_total rate — both miss the "had-duty-and-silently-dropped" case, and the coverage gauge isn't even exposed by every client.

A first-class counter for skips ends the guesswork and makes "missed aggregations" comparable across the fleet.

Proposed label values

reason	When it fires
not_aggregator	This slot's aggregation duty wasn't ours. Bookkeeping — lets you separate "no duty" from "had duty but skipped"
not_synced	Wall-lag or sync-status gate prevented aggregation (e.g. node is in behind_peers and aggregation is gated)
missing_state	Pre-state for the att_data target couldn't be resolved when the aggregator ran
spawn_failed	Aggregation worker queue was full / spawn error
other	Catch-all so clients can adopt incrementally without enumerating every internal failure mode

Sum across labels = total aggregation cycles seen. sum by (reason) (rate(lean_aggregator_skipped_total[5m])) then gives both the duty distribution and the genuine-miss rate.

Status table

Client	Status	Notes
Zeam	📝	Has equivalent counter under zeam_aggregate_skip_total; rename to lean_aggregator_skipped_total upstream-adoption
Others	□	Not yet implemented

Drive-by updates

lean_pq_sig_aggregated_signatures_building_time_seconds:

• Grandine: □ → ✅ — verified exposed on the live devnet (~600 observations on grandine_0 over a ~16 min run, p50≈1.19s)
• Zeam: 📝 → ✅ — implemented in zeam #941, exposed on devnet image sha256:bb801c18…, ~500 observations per aggregator (p50≈0.38s)

Test plan

• Reviewed metrics.md rendering locally
• Reviewers confirm naming + label set is acceptable
• Reviewers from each client team confirm/correct the status table