Clients SHOULD have unique metric IDs (name + tags)

If multiple applications/instances use the client and are scraped by the proxy application but they use the same metric names and tags, there will be duplicate time series in the Prometheus scrape (possibly with different values). Prometheus will only store one of these time series. The `instance` or `job` labels Prometheus adds will not help here because they will correspond to the Proxy application, not the Client application.

Start by locating how client metric names and tags are collected and exposed in the Prometheus scrape. Check whether multiple clients can produce identical time series, then determine the intended uniqueness boundary and define a regression test for the duplicate-series case. No file or test is named, so finding the relevant export path is part of the work.

Written by the indexing model from the issue text.