Track OpenTelemetry support at the Rust plugin boundary

[lucarlig]

Summary

This tracks the cpex-plugins portion of the OpenTelemetry/metrics work that was originally bundled into IBM/mcp-context-forge PR #4069.

That PR mixed changes across multiple repositories. This issue keeps the plugin-side work scoped to IBM/cpex-plugins so it can be implemented and reviewed independently once the framework provides the required standardized telemetry interfaces.

Blocked By

This issue is blocked on framework work in IBM/mcp-context-forge.

Before plugin implementation starts, the framework must provide a standardized way for all plugins to receive and return trace IDs and metrics data. That contract should define how trace context and metrics move across the host/plugin boundary, rather than each plugin inventing its own behavior.

This also depends on IBM/mcp-context-forge PR #3754 being integrated, followed by the framework implementation of the standard metrics path.

Scope for this repo

• implement plugin-side support for the standardized framework telemetry contract once available
• accept and propagate trace IDs through the Rust plugin boundary using that framework contract
• emit or return metrics through the standardized framework metrics path
• keep each plugin self-contained and avoid assuming an external OpenTelemetry collector/server or similar infrastructure is present
• cover plugin-side tracing and metrics behavior with focused Rust and Python integration tests where applicable

Out of Scope

• gateway/runtime changes that belong in IBM/mcp-context-forge
• defining the cross-plugin telemetry contract in this repository
• requiring an external OpenTelemetry server, collector, or similar service for plugin functionality
• cross-repo rollout coordination beyond linking companion issues

Background

• related to the work originally proposed in IBM/mcp-context-forge PR #4069
• blocked by IBM/mcp-context-forge PR #3754 and follow-up framework telemetry contract work
• created after the earlier cross-repo PR was closed so the work can be tracked per repository instead of as one cross-repo change

[lucarlig]

Blocked on framework work: we should wait for IBM/mcp-context-forge#3754 to be integrated, then wait for the framework to provide the standard way for plugins to pass trace IDs and metrics in and out. The framework-side protocol is tracked in contextforge-org/contextforge-plugins-framework#43. The plugin implementation should stay self-contained and must not assume an OpenTelemetry server, collector, or similar external service is available.