test(data): AUT-2246 use complete train state in MIMO loader test

[svcnemo-autobot]

Background

Launch_Unit_Tests_Core_2 failed after #5525 because its new canonical-grid test supplied an incomplete mock training state. The loader reads consumed_valid_samples, but the mock defined only consumed_train_samples.

What changed

Use the real TrainState fixture, whose train and validation sample counters both default to zero.

Details

This keeps the test aligned with the loader's declared TrainState contract and allows it to reach the canonical sampler behavior it intends to verify. Production behavior is unchanged.

Tested

• Launch_Unit_Tests_Core@2/3: 3,301 passed, 5 skipped
• Ruff check and format check on the changed file
• Full pre-commit could not run locally because the typed authoring checkout does not expose Git metadata inside the validation workstation; GitHub CI remains authoritative

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[svcnvidia-nemo-ci]

/ok to test c626c09