FLEET: Optional fork-like context inheritance for subagents

### Describe the feature or problem you'd like to solve

The current `/fleet` and subagent model is already very useful. Task-scoped delegation helps reduce noise and makes parallel work practical.

That said, I’d like an optional **fork-like** mode where a subagent can inherit a richer view of the parent task’s global context before focusing on a specific domain.

This would help in workflows where a generalist is solving a broad problem, temporarily switches into a specialist mode for one part of it, and then returns with a solution that is still aligned with the overall goal.

Today, task-scoped delegation can sometimes increase communication overhead or lead to small mismatches in goals, constraints, or architectural intent.

### Proposed solution

Add an optional context inheritance mode for `/fleet` subagents.

For example:

- `minimal`: current task-scoped behavior
- `balanced`: task context plus parent goals, constraints, and prior decisions
- `forked` / `full`: a richer inheritance of the parent task context before specializing

This would preserve the strengths of the current model while giving users a better option for subtasks that still need strong global awareness.

Conceptually, this is similar to a strong full-stack engineer working on a broad problem, temporarily switching into a more specialized mode for one part of it, and then returning with a result that is still aligned with the bigger picture.

The main benefit is reducing:

- communication cost
- goal-misalignment cost
- local optimizations that are correct in isolation but weaker in the broader task context

### Example prompts or workflows

1. Refactor an API across backend, frontend, tests, and docs while preserving backward compatibility.

2. Split a migration into schema, app, test, and doc subtasks, while keeping rollout and rollback constraints aligned.

3. Investigate a bug across multiple layers, delegate one area deeply, then merge the result back into the broader solution.

4. Tackle a large cross-cutting cleanup where each subtask still depends on shared architectural intent.

5. Work on a broad product or engineering task as a generalist, then temporarily delegate one part to a more domain-focused subagent without losing alignment with the parent task’s overall direction.

### Additional context

I’m **not** suggesting that the current `/fleet` behavior should change by default. I think the current `/fleet` and subagent model is good.

This request is specifically for an additional fork-like mode that helps when a subtask benefits from temporarily carrying more of the parent task’s full context.

A possible UX could be something like:

- `/fleet --context=minimal`
- `/fleet --context=balanced`
- `/fleet --context=forked`

or another mechanism that expresses the same idea.