Idea/Proposal: dynamic resctrl and perf activation metrics

Currently, perf and resctrl metrics are enabled either for all containers or none. Because hardware monitoring resources are limited, enabling them globally may not be practical on hosts with many containers.

Would the cAdvisor community be open to dynamically enabling these metrics for selected containers, through either:
- REST API endpoints to enable or disable perf and/or resctrl collection.
- A request parameter on fetch that activates the requested metrics on first use.

An explicit API may provide clearer lifecycle semantics, but we are open to suggestions. We would like to discuss the preferred approach before starting an implementation.

The issue names no files, tests, or entry points. First review how perf and resctrl metrics are currently enabled and how container metric requests are exposed, then clarify the preferred lifecycle and resource limits with the community. Done means an agreed API or request-parameter design with defined enablement and disablement semantics.

Written by the indexing model from the issue text.