metrics: add support for non-standard metrics

[yaauie]

This is an implementation detail of #18107, which will provide a ratio metric to track the bytes-in vs bytes-out of PQ compression.

As a: developer who needs to expose metrics via the Logstash HTTP API,
I want: a method for defining my own metric
So that: I can provide api-private metric implementations without wiring each new implementation through the public API

To this end:

Add a method to the public-api co.elastic.logstash.api.NamespacedMetric:

<U extends UserMetric<?>> U register(String metric, UserMetric.Factory<U> userMetricFactory);

Where UserMetric and UserMetric.Factory are defined as:

public interface UserMetric<T extends Serializable> {
    T getValue();

    interface Factory<U extends UserMetric<?>> {
        U create(String name);
        Class<U> getType();
        U nullImplementation();
    }
}

And wire it through to the underlying metric collector.

NOTE:

• naming TBD
• We may need a co.elastic.logstash.api.MetricType (which would be a super-interface for the existing enum org.logstash.instrument.metrics.MetricType) so that we can construct a org.logstash.instrument.metrics.Metric<?> -- although the MetricType is only actually used for gauge-type metrics.