Collect IntervalHistogram of monitorEventLoopDelay

The APM agent collects some runtime metrics. Event loop delay is one of them. When it's possible, the APM agent relies on the nodejs built-in functionality - [perf_hooks.monitorEventLoopDelay](https://nodejs.org/docs/latest-v16.x/api/perf_hooks.html#perf_hooksmonitoreventloopdelayoptions) method reporting event loop delay change over time. By default, `perf_hooks.monitorEventLoopDelay` returns [IntervalHistogram](https://nodejs.org/docs/latest-v16.x/api/perf_hooks.html#class-intervalhistogram-extends-histogram) object that groups data by value range buckets. However, the APM agent reports only a single value provided by the `IntervalHistogram` object - [mean](https://github.com/elastic/apm-agent-nodejs/blob/b7cf05a7d3d1b83af3eed86c2e234899815bcb7f/lib/metrics/runtime.js#L42-L43) value. It means that any deviations and spikes will not be reported. In general, this is not a problem, provided that the sample rate remains small (now it is [10 ms](https://github.com/elastic/apm-agent-nodejs/blob/b7cf05a7d3d1b83af3eed86c2e234899815bcb7f/lib/metrics/runtime.js#L13)). But if in the future we will collect metrics less often, the reported `mean` value will greatly skew the picture, smoothing out all the spikes and dips.

**Describe the solution you'd like**
There are a few different options. We could either report the whole histogram object, which comes with the cost of storing the additional amount of data. Or we can report a more aggressive metric to cover as many samples as possible (90th or 95th percentile, for example).