Agent Host OTel settings only work for Copilot, not Claude or Codex

[michiisai]

Turning on chat.agentHost.otel.enabled produces telemetry only for Copilot sessions. Claude and Codex sessions emit nothing.

Today only CopilotAgent requests a telemetry config from the Agent Host OTel service, the Claude and Codex agents don't. Each harness also needs a different approach: Claude's subprocess environment is deliberately minimal and drops OTel variables, and Codex reads its telemetry config from ~/.codex/config.toml rather than environment variables.

Asks:

• Emit the same core telemetry for Claude and Codex sessions. Copilot agent hose session currently produce:
  • gen_ai.operation.name
  • gen_ai.conversation.id
  • gen_ai.agent.name
  • gen_ai.request.model / gen_ai.response.model
  • gen_ai.usage.input_tokens / gen_ai.usage.output_tokens
  • gen_ai.usage.cache_read.input_tokens / gen_ai.usage.cache_creation.input_tokens
  • gen_ai.tool.name / gen_ai.tool.call.id
  • vscode.agent_host.session.uri
• Include a provider/harness attribute so consumers can tell sessions apart
• Make sure every agent host signal carries a consistent session identifier