SLURM_JOB_ID support for checkpoint output directory

**Is your feature request related to a problem? Please describe.**

If the same script is launched twice (i.e. with different hyperparams), the checkpoint output dir is not separated unless user manually changed the `checkpointing.checkpoint_dir`.

**Describe the solution you'd like**

I'd suggest an option to append `SLURM_JOB_ID` to `checkpointing.checkpoint_dir` same way as it's done for logs at ray.sub here https://github.com/NVIDIA-NeMo/RL/blob/main/ray.sub#L98

**Describe alternatives you've considered**

One way is to have an env var like `CKPT_DIR_SUFFIX` which causes ray.sub to append smth like "10001-run/" to `checkpointing.checkpoint_dir`. It is quite hardcoded though.

Another way would be to support interpolations with key words in nemo-rl configs, i.e. if user config has `checkpointing.checkpoint_dir: output/ckpts/${slurm_job_id}-run`, it will be replaced with `SLURM_JOB_ID` at ray.sub stage. That would take more time to implement.

The downside is that ray.sub will mess with user defined args which is not the case at the moment

**Additional context**

torchtune does interpolations -> https://docs.pytorch.org/torchtune/stable/deep_dives/configs.html#referencing-other-config-fields-with-interpolations but without key words