fix(framework): Stop logging executors before completion

[tanertopal]

What changed

• stop heartbeat workers before task completion in AgentApp, ServerApp, and Simulation executors
• drain and stop log uploaders within the same cleanup budget
• send final task output with only the remaining exit-handler time
• close the runtime connection after the final output attempt

Why

These executors have two authenticated background workers. Both must stop before task completion revokes the token, but their sequential cleanup must not consume the time reserved for final task output.

sequenceDiagram
    participant E as Executor
    participant H as Heartbeat worker
    participant L as Log uploader
    participant R as Runtime API
    E->>E: Disable retries
    E->>H: Stop
    E->>L: Drain and stop
    Note over H,L: Shared 2.5s cleanup deadline
    E->>R: PushTaskOutput with remaining deadline
    E->>E: Close transport

Loading

Stack

Umbrella/reference: #7895 — Prevent task executor shutdown deadlock

• Previous: #7919 — Stop heartbeat executors before completion
• This PR: fix(framework): Stop logging executors before completion #7920 — Stop logging executors before completion
• Next: #7921 — Preserve output before runtime cleanup

Merge the stack in order. After the previous PR merges, retarget this PR to main.

Validation

• pytest py/flwr/superlink/runtime/run_serverapp_test.py py/flwr/simulation/app_test.py
• Ruff and Black on the changed executor modules
• git diff --check