feat(node): Add garbage collection metrics to NodeRuntimeMetrics

### Description

`NodeRuntimeMetrics` already collects heap, RSS, event loop delay, and ELU, but has no visibility into GC — often the direct cause of event-loop stalls and tail latency.

Add GC pause metrics collected via a `perf_hooks` `PerformanceObserver` on `gc` entries, aggregated over the existing collection interval:

- `gc.pause.duration` (total/avg pause time)
- `gc.pause.max` or `gc.pause.p95` (worst pause in the window)
- `gc.count` broken down by kind (minor/scavenge vs major/mark-sweep)

Opt-in by default, consistent with the other advanced metrics in the `collect` option. Aligns with the OTel `process.runtime.v8js.gc.*` conventions.

Start at the NodeRuntimeMetrics implementation and its existing collection interval and advanced collect options. Read the Node.js perf_hooks PerformanceObserver documentation for gc entries, then trace how current heap, RSS, event-loop delay, and ELU metrics are aggregated. Done means GC pause totals and an aggregate worst-pause metric are reported, with counts split by minor and major collection kind.

Written by the indexing model from the issue text.