fix: Add timeout to IMDS requests to prevent shutdown hang

[folouiseAWS]

What was the problem/requirement?

The IMDSv2 token request in Worker._get_ec2_metadata_imdsv2_token() (and the spot instance-action and ASG lifecycle-state requests) had no timeout and only caught requests.ConnectionError.

On hosts where the IMDS address (169.254.169.254) is silently dropped rather than actively refused — possible on any non-EC2 host depending on network configuration — the token request in Worker.run() blocks the main thread indefinitely. The scheduler keeps running in the executor, so the agent appears healthy and processes work, but SIGTERM shutdown can never complete: the worker is never transitioned to STOPPED and must be SIGKILLed, leaving the service-side worker resource in STARTED until it ages out to NOT_RESPONDING.

What was the solution?

• Add timeout=0.5 to all three IMDS requests in worker.py, matching the timeout and rationale already used by startup.bootstrap._get_metadata.
• Treat requests.Timeout the same as requests.ConnectionError: IMDS is unavailable.
• Extracted as a class constant Worker._IMDS_REQUEST_TIMEOUT_SECONDS with documentation.

What was the testing done?

• Unit tests added for timeout behavior of all three IMDS queries; existing call assertions updated. Full test/unit/test_worker.py module passes (36 tests).
• hatch run lint (ruff + mypy) passes.
• Reproduced live on an arm64 macOS host on a network where the IMDS address is silently dropped, running the agent against a real Deadline Cloud farm: without this change, the agent hung indefinitely after SIGTERM (thread dump showed the main thread blocked in requests.put at worker.py:391) and the worker stayed STARTED; with this change, the agent transitioned the worker to STOPPED and exited in under one second.