pubsub: add CPU utilization tracking and idle stress scenario to throughput benchmark

This issue tracks adding a dedicated resource utilization benchmark tool (e.g. `pubsub-cpu-benchmark` under `src/pubsub/benchmarks/`) to measure client CPU and memory footprint across idle and active phases.

As demonstrated in PR #6050 (https://github.com/googleapis/google-cloud-rust/pull/6050#discussion_r3590569283 and https://github.com/panimidi/pubsub-demo), measuring client process CPU utilization (`getrusage`) when spawning thousands of publishers—both at rest and under load—is critical to prevent CPU regressions.

We should include a benchmark like this to prevent regressions in the future.

Start by reviewing the existing benchmarks under src/pubsub/benchmarks/ and the discussion in PR #6050. Define the proposed pubsub-cpu-benchmark around client CPU and memory measurements during idle and active phases, including thousands of publishers and getrusage-based CPU tracking. Done means the benchmark can run both scenarios and provide results suitable for detecting regressions.

Written by the indexing model from the issue text.