feat: add speed metric

[declanjackson]

Add Model Speed Metrics

Summary

• Add speed and timing metrics to message models for tracking LLM and tool performance
• AssistantMessage now has request_start_time/request_end_time fields and a derived e2e_otps (end-to-end output tokens per second)
  property
• ToolMessage now has tool_start_time/tool_end_time fields and a derived tool_duration property
• All three clients (ChatCompletions, LiteLLM, OpenResponses) record request timing
• Agent records tool execution timing and aggregates model speed stats per run
• Logger displays a Model Speed summary panel (model, OTPS, calls, gen time) and average tool duration in the tool summary

Other Changes

• Rename TokenUsage.output to TokenUsage.answer to clarify token terminology (output = reasoning + answer); output remains as a deprecated
  computed property for backwards compat
• Add model_slug to the agent run summary
• Improve summary panel layout to adapt to terminal width instead of forcing a single row
• aggregate_metadata skips internal _-prefixed keys in run_metadata
• Add guard for None shared exec_env in sub-agent code path

[MohamedAkbarally]

LGTM