Add cancellable execution contexts, deadlines, budgets, and progress events

Problem

The processor interface is a synchronous StartProcessing() call with an atomic percentage value. It has no cancellation token, deadline, resource budget, checkpoint, structured progress event, or partial-result callback.

This makes long GCS, inference, compute, and render jobs hard to stop safely, monitor by pass, or limit when a branch is no longer needed.

Scope

Introduce an execution context passed through ProcessingManager and every executor with:

• Execution, attempt, pass, and subtask identifiers.
• Cooperative cancellation token.
• Overall and per-pass deadlines.
• CPU, GPU, RAM/VRAM, disk, network, runtime, and output-size budgets where enforceable by the executor.
• Structured progress events with pass ID, stage, completed work, total work, and message.
• Partial-result and checkpoint callbacks for processors that support them.
• Clear terminal states for completed, cancelled, timed out, budget exceeded, and failed.
• Safe cleanup of Vulkan resources, MNN sessions, loaded artifacts, and pending asynchronous saves.

Do not duplicate SuperGenius queue locking or task ownership. This issue defines the SGProcessingManager execution-control contract that SuperGenius can drive.

Acceptance criteria

• A running MNN or Vulkan test job can be cancelled without leaking resources or publishing a successful result.
• Deadline and output-size limits produce distinct typed failures.
• Progress events identify the active pass and stage rather than only exposing one float.
• Checkpoint-capable tests can resume from a saved checkpoint; processors without checkpoint support report that clearly.
• Existing processors remain usable through an adapter during migration.

Supports GCS early cancellation and long-running compute/render workloads.

Base implementation and review work on the branch consumed by SuperGenius/develop.