test: unskip nova tests

[lucasjia-aws]

Summary

Enable Nova model integration tests to run in the dedicated us-east-1 test account (784379639078). This PR migrates the Nova tests from being skipped to being selectable via a new us_east_1 pytest marker, allowing them to run in the new account's CI infrastructure set up in #5900.

Changes

• New us_east_1 pytest marker (tox.ini): Registered a new marker to identify tests that require us-east-1 account credentials.
• Unskip Nova tests: Replaced @pytest.mark.skip with @pytest.mark.us_east_1 on all Nova-related integration tests.
• Update resource references: Updated S3 paths, ARNs, and account IDs from the old account (729646638167) to the new dedicated test account.
• Extend timeouts: Increased max_wait_time to 3 hours for SFT/RLVR trainer tests and benchmark evaluator test, as Nova training jobs take >1 hour to complete.

Tests affected

Test	File	Markers
test_sft_trainer_nova_workflow	test_sft_trainer_integration.py	gpu_intensive, us_east_1
test_rlvr_trainer_nova_workflow	test_rlvr_trainer_integration.py	gpu_intensive, us_east_1
test_benchmark_evaluation_nova_model	test_benchmark_evaluator.py	gpu_intensive, us_east_1
test_create_dataset_from_s3_nova_* (4 tests)	test_dataset.py	serial, us_east_1

How it works

• PR checks (us-east-1 CodeBuild): Runs -m "us_east_1 and not serial" and -m "us_east_1 and serial" via the integ-tests-us-east-1 job added in test: add a job to invoke code build project in us-east-1 #5900.
• Scheduled GPU tests: The gpu_intensive marked tests (SFT, RLVR, benchmark) will run in the scheduled gpu-integ-tests-us-east-1 job.
• Existing us-west-2 tests: Unaffected. The us-west-2 integ test job should add -m "not us_east_1" to exclude these tests.

Testing

All 7 tests validated via PR check runs

[aviruthen]

Are we running these tests in a different account? E.g. 784379639078 instead of 729646638167?

[lucasjia-aws]

yes, @mujtaba1747 set that up earlier

[mujtaba1747]

There's no cleanup for DataSet.create()?

[lucasjia-aws]

There is cleanup — the cleanup_list fixture handles it. Each test appends the created dataset to cleanup_list, and the fixture's teardown calls resource.delete() on all tracked resources (see conftest.py line ~108-113).