stats: add an adapter sink that forwards to Envoy stats sinks

[bpalermo]

Description

Adds nighthawk.envoy_stats_sink_adapter, a Nighthawk stats sink plugin that forwards to a stats sink implemented as an Envoy extension, such as envoy.stat_sinks.statsd or envoy.stat_sinks.dog_statsd.

What is new:

• The plugin (source/stats_sink/) and its config proto (api/stats_sink/), which nests an ordinary Envoy StatsSink config.
• A createStatsSink(config, context) overload on NighthawkStatsSinkFactory. Envoy's sink factories need the sink configuration and a server factory context, and the existing method receives neither.

What stays the same:

• statistic_impl.cc and statistic_test.cc are not touched.
• The new overload's default implementation calls the existing method, so existing stats sink plugins behave as before. process_test, which runs a Nighthawk sink through ProcessImpl, passes unchanged.

The adapter forwards flushes unchanged. Nighthawk records latency samples in nanoseconds with no unit declared; the adapter forwards them as microseconds with that unit, named cluster.<worker>.<statistic>, so sinks that scale by unit report them correctly.

This PR is related to #1616 and #1617, which it replaces.

Notes for Reviewers

• Tests: //test/stats_sink:envoy_stats_sink_adapter_test (unit, with a fake Envoy sink) and test/integration/test_stats_sinks.py, which runs Envoy's real statsd and DogStatsD sinks through the adapter and checks the datagrams received on a UDP socket.
• Limitation: Nighthawk's latency statistics are not in the flush snapshot, so Envoy sinks that only read histograms from the snapshot (OpenTelemetry, metrics service) export counters and gauges but not latencies. Documented in docs/root/statistics.md.