feat: Adapt eigent Workforce to CAMEL's plan-then-run API

[fengju0213]

Adapt eigent Workforce to CAMEL's plan-then-run API with a thin local planning adapter

Summary

Replace eigent's duplicated orchestration core with CAMEL's new plan_task_async() / run_plan_async() APIs, while keeping eigent-specific planning context injection as a thin local adapter. This removes most of the manual queueing/sync logic without pushing product-specific semantics upstream into CAMEL.

Changes

backend/app/utils/workforce.py

Removed or simplified	Replaced by
Manual decomposition/execution orchestration	Delegates to CAMEL plan_task_async() / run_plan_async()
Manual pending queue and snapshot startup logic	Uses CAMEL plan/run lifecycle
Manual _sync_subtask_to_parent logic	Uses CAMEL _sync_task_to_parent behavior
Large handle_decompose_append_task override	Deleted in favor of upstream planning flow
eigent-specific planning context behavior	Kept as a thin eigent-side adapter instead of becoming CAMEL public API

Backward compatible:

• eigent_make_sub_tasks() still returns list[Task] and stores the latest WorkforcePlan internally
• eigent_start() still accepts both WorkforcePlan and edited list[Task]
• Existing frontend notification overrides remain unchanged

Why this split

What CAMEL now owns:

• public plan-then-run orchestration
• task tree consistency guarantees
• neutral event/state metadata

What eigent continues to own:

• planning-only context injection for its coordinator workflow
• frontend notification payloads and mapping logic
• product-specific retry, timeout, and cleanup behavior

This keeps CAMEL minimal and reusable, while still letting eigent delete most of its duplicated workflow code.

[fengju0213]

pending： camel-ai/camel#3979