improve traces correlation

**Background**
Traces that are generated during pipeline run are correlated by pipeline run id only in case of `pipeline.run` method. Otherwise steps from the same pipeline and same script are not correlated.

**Tasks**
1. Correlate traces with "runtime run id" which correlates all pipelines running in particular Python interpreter instance (ie. script execution). Make it configurable via `RUNTIME__RUN_ID` (RuntimeConfiguration) so users can control it.
2. Correlate traces from a particular instance of pipeline (pipeline_instance_id)
3. improve pipeline run id correlation
- `pipeline.run` - all steps executed in that method are correlated
- `pipeline.extract` and other steps called explicitly: correlated under ad hoc run
4. make `pipeline.last_trace` to contain last pipeline run: explicit or ad hoc
5. document it in `running in production`

Start by locating RuntimeConfiguration and the implementations of pipeline.run, pipeline.extract, and pipeline.last_trace. Trace how run IDs and pipeline instances are currently assigned and find the relevant tests before changing behavior. Done means the listed correlation cases work, RUNTIME__RUN_ID is configurable, last_trace covers explicit and ad hoc runs, and the running-in-production documentation is updated.

Written by the indexing model from the issue text.