feat(telemetry): include skill_name in safe tool metadata whitelist #30155

What would you like to be added?

Add skill_name to TOOL_CALL_METADATA_SAFE_KEYS and emit it within the execution result of ActivateSkillToolInvocation.

When enterprise environments operate with telemetry.logPrompts: false (to comply with corporate security and privacy policies forbidding prompt logging, raw arguments, shell scripts, and PII), the OpenTelemetry exporter strips all tool call parameters. As a consequence, telemetry captures that a tool was called, but cannot identify which Agent Skill was activated.

Why is this needed?

Unlike free-form user prompts or arbitrary shell arguments, a skill name is a safe, catalog-defined identifier (e.g., skill-creator, verification-loop) that carries zero risk of leaking confidential data, source code, or PII.
Allowing skill_name in the safe metadata allowlist enables organizations and engineering leaders to observe Agent Skill adoption, frequency, and tool performance metrics across enterprise teams without compromising user privacy or violating data residency/security guardrails.

Additional context

A complete implementation with passing unit tests is already prepared and submitted in PR #30010.

Could a maintainer please review this proposal and, if aligned with the project roadmap, add the help wanted label so the pull request can be considered?
Thank you!