feat(tasks): add run_network_disruption task and rewrite blackout playbook

[parithosh]

What

Adds a native run_network_disruption task that drives a disruptoor instance declaratively from playbooks — the task type planned as step 3 of disruptoor's design doc — and rewrites playbooks/dev/single-node-blackout-recovery.yaml (#218) on top of it.

- name: run_network_disruption
  title: "Black out the target beacon node"
  config:
    disruptoorUrl: "http://disruptoor:7700"
    isolations:
      - name: blackout-target-cl
        target: { node-index: 1, client-type: beacon }
        scope: [cl_p2p, el_p2p, include_control]

- name: run_network_disruption
  config: { disruptoorUrl: "http://disruptoor:7700", action: clear }

Task design

• Actions: set (replace the whole disruptoor state), update (read-merge-write: replaces entries by name, appends new ones, drops names in removeNames; guarded by If-Match and retried on concurrent writes — lets playbooks compose disruptions, e.g. keep a baseline jitter while toggling a blackout), clear (heal everything).
• Passthrough entries: partitions / isolations / shaping go to disruptoor verbatim (wire format of PUT /v1/state), so new disruptoor fields never require assertoor changes. Assertoor only enforces unique non-empty entry names (needed for merge); disruptoor's precise validation errors are surfaced in the task failure.
• Built-in API wait: awaitApiTimeout / pollInterval poll healthz before acting, replacing sleep + curl shell tasks.
• Outputs: appliedState (read back via GET /v1/state, which reflects applied reality) plus partitionCount / isolationCount / shapingCount.

Playbook rewrite

Supersedes the run_shell/curl/jq version from #218: the two-partition workaround and the jq counterparty enumeration (with its totalCount output dependency) collapse into a single disruptoor isolation, whose counterparty group disruptoor computes as the complement of the target — 193 → 135 lines, zero shell, works unchanged on any topology. A leading action: clear task doubles as a fail-fast health check and stale-state cleanup. Runtime semantics are identical to what was validated on glamsterdam-devnet-7 in #218: the target CL loses all traffic including to its own EL/VC, while the target's EL keeps its network peers.

Dependencies

• Requires a disruptoor build including the isolations primitive (feat: add isolations primitive to v1 state disruptoor#13); older disruptoor versions reject the unknown field with a clear 400.
• Once this lands, feat(playbooks): add single-node blackout & recovery playbook #218 can be closed in favor of this playbook (or kept if a shell-only fallback for older disruptoor images is wanted).

Testing

• go build ./... and the full go test ./... suite green; golangci-lint run --new-from-rev=origin/master reports 0 issues.
• New unit tests (21 cases): config validation table, ETag 412 retry, update merge/remove semantics, server-error surfacing, healthz probe, and a test exercising the exact playbook pattern of building isolations via a configVars jq expression through the vars system.
• Playbook passes assertoor validate; playbook index regenerated via make generate-playbook-index.

🤖 Generated with Claude Code