fix: flatten the pages in DeploymentResourcePool.query_deployed_models - #7231

Fixes #7230 🦕

DeploymentResourcePool.query_deployed_models() is documented, and annotated, to return a list of DeployedModelRef objects, but it built that list with itertools.chain(page.deployed_model_refs for page in response.pages). Passing a single generator to chain does not flatten it, so the result held one RepeatedComposite per page instead of the refs themselves. With one page holding two refs the method returned a list of length 1, and refs[0].deployed_model_id raised AttributeError.

This switches to itertools.chain.from_iterable(...) in both the GA class (google/cloud/aiplatform/models.py) and the preview class (google/cloud/aiplatform/preview/models.py), so the method returns what its docstring and return type promise.

Tests in tests/unit/aiplatform/test_deployment_resource_pools.py:

• test_query_deployed_models now checks the returned deployed_model_id values, not only the length. The length check passed with the bug because the fixture has two pages with one ref each.
• test_query_deployed_models_with_several_refs_in_one_page covers one page holding two refs.
• test_preview_query_deployed_models covers the preview class.

All three fail on main and pass with this change, and the whole file passes (12 passed). black==24.8.0 and flake8==6.1.0 from the noxfile report nothing on the changed files.

With the script from #7230:

main (faed3f4)
len: 1 | item types: ['RepeatedComposite']
refs[0].deployed_model_id -> AttributeError: 'RepeatedComposite' object has no attribute 'deployed_model_id'

this branch
len: 2 | item types: ['DeployedModelRef', 'DeployedModelRef']
first deployed_model_id: m1

• Issue opened with a reproduction (DeploymentResourcePool.query_deployed_models returns page containers instead of DeployedModelRef objects #7230)
• Tests and linter pass
• Code coverage does not decrease
• No docs change needed, the docstring already describes the fixed behavior