Implement average lifetime long batch's size and document count metric

[andsel]

Create the metrics partial subtree related to batch.event_count.average.lifetime and batch.byte_size.average.lifetime.
This initial implementation must also expose a setting named pipeline.batch.metrics.sampling_mode which defines 3 behaviours:

• disabled: no collection of any batch data, so the response of HTTP API shouldn't change respect to actual shape.
• minimal: collect the metrics for a subset of all the batches (1% of the total).
• full: collect the metrics data from each batch.

Initially this setting is none by default and is defined per pipeline.

Consider the work study PR #17758 for the core part of memory occupation estimation.

• expose the event_count for memory and persistent
• expose the byte_size for memory and persistent
• implement the flag