Add opt-in STAGGERED onboarding mode

[ShvaykaD]

Summary

Adds an opt-in STAGGERED onboarding mode for the persistent gateway and device tests. Instead of the phased warmup (connect-all → subscribe-all → announce-all), each entity runs its own lifecycle — connect → announce sub-devices (gateway) → subscribe → per-entity telemetry — paced by a concurrency cap and startup jitter. When all entities reach a terminal state (onboarded or failed), the existing RPC burst sender starts unchanged on the full device list.

Motivation: the phased warmup produces a synchronized connect/announce storm at startup. Staggering onboarding is closer to a real device fleet coming online gradually and avoids the artificial cold-start spike, while keeping the same steady-state load so results stay comparable to PHASED.

Opt-in / backward compatibility

• New ONBOARD_MODE config key; default PHASED = existing behavior, unchanged.
• Every new behavior is gated on ONBOARD_MODE=STAGGERED. The PHASED, ephemeral, and device-default paths are unchanged.

Config keys

Key	Default	Purpose
ONBOARD_MODE	PHASED	PHASED (existing) or STAGGERED
ONBOARD_MAX_CONCURRENT	200	max entities onboarding at once
ONBOARD_FIRST_JITTER_SEC	60	spread of per-entity onboarding start times

Steady-state matches PHASED: per-entity telemetry period = entityCount / MESSAGES_PER_SECOND seconds; no telemetry timer is scheduled when MESSAGES_PER_SECOND <= 0.

Scope / known limitations

• Gateway STAGGERED requires gateway.batch=true and no alarms; it fails fast on other combinations rather than silently diverging.
• Device STAGGERED is telemetry-only (there is no device-RPC path in the tool).
• No live CONNECTIONS gauge under STAGGERED. Throughput and RPC/announce stats print periodically, and end-of-run totals are reported.

Tests

• New unit tests: onboarding engine (concurrency cap ≤ max, ramp-complete fires exactly once, per-entity failure counting), per-entity telemetry cadence derivation, commit-on-success only (a mid-onboard failure leaves no entry in the shared collections / RPC target list), and PHASED vs STAGGERED device-assignment parity.
• Full suite green: 178/178 via mvn -T 1C clean install.

See STAGGERED-ONBOARDING-RUNBOOK.md for a usage/verification runbook.