fix: clean up adaptive timeout signals in connection monitor

[lodekeeper]

Context

Fixes a steady memory growth path in heartbeat pinging.

Related issue: #3391

What this changes

1) Always clean up connection-monitor timeout signals

ConnectionMonitor now wraps each heartbeat ping attempt with:

• const signal = this.timeout.getTimeoutSignal(...)
• finally { this.timeout.cleanUp(signal) }

This ensures cleanup runs on success, timeout, and error paths.

2) Ensure AdaptiveTimeout.cleanUp releases composed signal listeners

AdaptiveTimeout.cleanUp(signal) now calls:

signal.clear()

before computing/recording timing stats.

This detaches listener chains created by anySignal(...)/AbortSignal.timeout(...).

Tests

Added tests:

• packages/utils/test/adaptive-timeout.spec.ts
  • verifies cleanUp calls signal.clear()
• packages/libp2p/test/connection-monitor/index.spec.ts
  • verifies ConnectionMonitor invokes timeout cleanup during heartbeat

Notes

This patch was validated downstream with a long A/B soak where unpatched nodes grew ~2 MB/h faster than patched under identical conditions.

[dozyio]

Thanks @lodekeeper !