[barnabasbusa]

check_consensus_attestation_stats only had network-wide thresholds, so a client pair whose validators stopped voting hid under the overall percentage when the other pairs carried participation, which is exactly what client interop runs need to catch. This adds validatorNamePattern to restrict the duties and votes to matching validators (same idea as in check_consensus_block_proposals), so a run_task_matrix over validatorPairNames checks every pair against its own target and head thresholds, and makes the progress and failure messages say which threshold missed, e.g. epoch 14 failed: target votes 94.00% < 98%, instead of a bare "check failed". Verified on a local Lodestar/Prysm crossed devnet.

Examples

Check every client pair against its own thresholds:

- name: run_task_matrix
  configVars:
    matrixValues: "validatorPairNames"
  config:
    runConcurrent: true
    matrixVar: "validatorPairName"
    task:
      name: check_consensus_attestation_stats
      title: "Votes of ${validatorPairName}"
      configVars:
        validatorNamePattern: "validatorPairName"
      config:
        minTargetPercent: 95
        minHeadPercent: 90
        minCheckedEpochs: 3

Only the validators of one pair:

- name: check_consensus_attestation_stats
  config:
    validatorNamePattern: "^3-"
    minTargetPercent: 98
    failOnCheckMiss: true

Progress and failure messages now name the missed threshold:

before: Waiting for attestation stats... (attempt 4)
        attestation stats check failed for epoch 14
after:  Waiting for attestation stats, epoch 14 failed: target votes 94.00% < 98% (attempt 4)
        attestation stats check failed for epoch 14: target votes 94.00% < 98%
        attestation stats check failed for epoch 14: no active validators match

[redpandabot]

Adds a validatorNamePattern regex to check_consensus_attestation_stats that scopes both the active-balance denominator and the counted attestation votes to matching validators, plus human-readable per-threshold failure reasons. The filtering is applied carefully (duties keep all validators with a matches flag so Electra aggregation-bit offsets remain aligned), is backwards compatible when no pattern is set, and guards the previous 0/0 division. I found no real defects.

---

_{Reviewed 5 changed file(s) @ 5db2912f — no blocking issues found.
"Hope is not a strategy." — Google SRE}