Cardinality for Elasticsearch span names is too high

Currently, the span name is `Elasticsearch: ${http.method} ${http.url.path}`. The issue is that the path can contain document IDs, such as `GET /customers/_doc/42`.

This is problematic if we want to roll up metrics based on `span.name`.

Agents may only have the access to the HTTP request info that the low-level RestClients expose.

An idea that needs further validation is to remove any path segments that come after a path segment that begins with an underscore. Examples:
* `GET /customers/_doc/42` -> `GET /customers/_doc`
* `GET /_cluster/health/my-index-000001` -> `GET /_cluster`
* `GET /_alias/my-alias` -> `GET /_alias`

While this strategy might sometimes result in what feels like too low cardinality (`GET /_cluster/health/my-index-000001` or `GET /_cluster/health` is not problematic), it's an easy strategy that should work quite generically.

As this would chop off some parts of the URL, it becomes important for agents to collect the full URL in `context.http.url`. This overlaps with https://github.com/elastic/apm-agent-nodejs/issues/2019

While at it, we should also consider dropping the `Elasticsearch: ` prefix as suggested here: https://github.com/elastic/apm/pull/420#issuecomment-828279997