fix(crew): deduplicate manager agent token usage accounting

[wishhyt]

Summary

• calculate_usage_metrics() counted the manager agent's token usage twice: once via _token_process.get_summary() and again via llm.get_token_usage_summary(), because the two checks were independent if blocks instead of mutually exclusive branches.
• For regular agents, the code correctly uses if/else to choose one path. The manager agent path was missing the else.

Changes

Restructured to use the same if isinstance(llm, BaseLLM) / else pattern as regular agents.

Test plan

Run a hierarchical crew and verify usage_metrics values are not inflated

---

Note

Low Risk
Low risk: small, localized change to usage metrics aggregation logic that only affects reported token counts for hierarchical crews.

Overview
Fixes Crew.calculate_usage_metrics() so the manager agent’s token usage is only added once by making the BaseLLM.get_token_usage_summary() path and the _token_process.get_summary() fallback mutually exclusive (matching the regular agent logic). This prevents inflated usage_metrics/token totals when both mechanisms are present on the manager agent.

^{Written by Cursor Bugbot for commit d2334c2. This will update automatically on new commits. Configure here.}

[github-actions]

This PR is stale because it has been open for 45 days with no activity.

[Vidit-Ostwal]

Closing this. The manager if/else only stops one double-count while still reconstructing usage_metrics from LLM lifetime totals (and _token_process). That source is wrong for shared instances, leftover usage from a prior run, and the manager path.

Direction: per-agent UsageMetrics via snapshot/delta around each agent run, then calculate_usage_metrics() sums those. Kickoff already does the snapshot for LiteAgentOutput; persist it on the agent (not TokenProcess) and include the manager the same way.

Follow the review on #7260 for that rework. This PR is not the shape we want to land.