scripts/iris/dev_tpu.py still calls removed IrisConfig.platform()

[yonromai]

Describe the bug

scripts/iris/dev_tpu.py still calls the removed IrisConfig.platform() API in controller_client(...) (scripts/iris/dev_tpu.py:172). Current IrisConfig no longer exposes platform(); it now exposes provider_bundle() and controller_address() (lib/iris/src/iris/cluster/config.py:1050, lib/iris/src/iris/cluster/config.py:1067).

Once local Iris imports are otherwise healthy, the dev TPU workflow crashes before allocation completes with:

AttributeError: 'IrisConfig' object has no attribute 'platform'

To Reproduce

1. Start from current main.
2. Ensure Iris protobuf bindings are current so the script gets past import-time failures.
3. Run:

uv run python scripts/iris/dev_tpu.py \
  --config lib/iris/examples/marin.yaml \
  --tpu-name repro-dev-tpu \
  allocate --tpu-type v5p-8

4. Observe the script crash on iris_config.platform().

Expected behavior

dev_tpu.py should use the current IrisConfig API and reach real controller communication instead of failing on an attribute lookup.

Additional context

The stale caller comes from cb6da22fd / #3546 (Add Iris dev TPU workflow). The API removal happened in 1e1d2dbd23 / #3900 (iris: eliminate Platform abstraction, reorganize into Service + Provider layers). A minimal fix is to use iris_config.provider_bundle().controller and then call discover_controller(...) / tunnel(...) on the controller provider.