test(extras): add an end-to-end test for the microvm operation

[Vivek0712]

Stacked on #769: this PR targets feat/microvm-operation, so the diff shows only the e2e commit. It moves to main when #769 merges.

This ports the JavaScript SDK's extras/e2e to Python, as packages/aws-durable-execution-sdk-python-extras/e2e/run_e2e.py. It runs the microvm operation and the MicroVM worker against a real account. Both are built from the branch, so a run tests the change under review.

What it does

1. Builds the core SDK, extras, and the MicroVM worker as wheels.
2. Creates or reuses a bucket and three roles, all named dex-microvm-e2e-py-*.
3. Builds a MicroVM image from e2e/microvm/ with the worker. The image name carries a hash of the sources, so an unchanged worker reuses the image.
4. Deploys e2e/function/index.py as a durable function, Python 3.13 on arm64, and publishes a version.
5. Runs each scenario as a durable execution, and checks the outcome, the operation history, and that every launched MicroVM ends TERMINATED.

The scenarios and the function's events are copied from the JavaScript runner. So a case added on one side can be added on the other with the same event.

Scenarios

Default run: succeed, fail, crash, too-large-for-the-run-hook, run-hook-cjk, run-hook-emoji, parallel. Optional: long-job.

Pending, skipped unless named in E2E_SCENARIOS:

• terminated: needs the worker's terminate action.
• request-succeed, request-fail, auto-http, auto-http-cjk: need delivery over HTTP. too-large-for-the-run-hook checks today's behaviour and goes away then.
• session, suspend-session, idle-probe: need microvm_session.

Each pending scenario already has the JavaScript expectations. Turning one on means removing pending and adding its branch to the function.

Differences from the JavaScript runner

• It uses boto3 instead of the AWS CLI.
• It compares the error type's class name. Python reports the fully-qualified name.
• It expects the core subtypes (RunInChildContext, Callback, Step), since custom subtypes wait for [Feature]: Custom subtype for step, callback, and child context operations #749. E2E_SUB_TYPES=custom switches to the JavaScript ones.
• It starts a scenario only when its MicroVMs fit E2E_MAX_MICROVMS, 4 by default. A new account may allocate 8 GB of MicroVM memory, and each MicroVM has 2 GB. Without the budget, parallel hit ServiceQuotaExceededException on one item, and the other item's MicroVM kept running.
• E2E_IMAGE_ARN and E2E_FUNCTION_ARN reuse an existing image or function. With the JavaScript e2e image, terminated passes, which checks the operation against both workers.

Testing

Run in us-east-1 from this branch:

• Default run: 7 of 7 passed, every MicroVM TERMINATED.
• With E2E_IMAGE_ARN set to the JavaScript e2e image: succeed and terminated passed, the latter with MicrovmJobFailedError (MicrovmTerminatedError) 20 s in.
• ruff check, ruff format --check, and mypy pass on the new files.

Wiring it into cloud-tests.yml needs the test account's image build role, so I left that out.

🤖 Generated with Claude Code