Upgrade ai ml tests to successfully run for long duration on another VM

[sethiay]

Description

Upgrade ai ml tests to successfully run for long duration on VM other than the VM on which the kokoro tests are triggered. This is required because of intermittent issues that comes while running long duration tests on Kokoro VM directly.

Notes for reviewer:

1. I have used sudo gcloud instead of gcloud because gcloud is giving errors while doing ssh from Kokoro VM and I couldn't find a solution.
2. I have changed the base image for Pytorch dino model because the nvidia one doesn't have gsutil preinstalled.
3. I will take care of to-dos before merging.
4. To-do: Take care of deleting logs for older build in GCS buckets, similar to periodic perf tests.
5. The existing build.sh files for both tf and pytorch are moved under scripts/ml_tests/tf/resnet/ and scripts/ml_tests/pytorch/dino by the name setup_host_and_run_model.sh

Link to the issue in case of a bug fix.

NA

Testing details

1. Manual - Tested the setup for tf model with less number of epochs. To-do/In progress: Test the setup for pytorch and with relatively higher number of epochs.
2. Unit tests - NA
3. Integration tests - NA