Lazy resampling: MetaTensor.applied_operations ends up out of order

Describe the bug
When executing pipelines with lazy resampling, the applied_operations list on a MetaTensor instance can end up out of order.

To Reproduce
Steps to reproduce the behavior:

1. Run tests/test_integration_lazy_samples.py
2. Add a breakpoint at https://github.com/Project-MONAI/MONAI/blob/4f291723c4ce1a1baf7efbd499495b777dfb1869/tests/test_integration_lazy_samples.py#L155
3. Examine the state of batch_data and you'll see that the applied operations are out of order. Lambdad is the last transform in the pipeline, but appears as the first element in the pipeline

Correct order:

[SpatialResample, Orientation, RandRotate90d, RandCropByPosNegLabel, RandRotated, RandZoomd, ResizeWithPadOrCrop, Rotate, Lambdad]

Observed order:

[Lambdad, SpatialResample, Orientation, RandRotate90d, RandCropByPosNegLabel, RandRotated, RandZoomd, ResizeWithPadOrCrop, Rotate]

Note that this issue is fixed in #6257

Start with tests/test_integration_lazy_samples.py at line 155 and inspect batch_data during the lazy resampling pipeline. Compare the observed applied_operations order with the expected list in the issue, then review the referenced fix in #6257; done means the integration test shows operations in pipeline order.