GPU selection remains even after GPU nodes are gone

Description of problem:

• A non-GPU pod can no longer be spawned because the UI remembers the last choices made.

Prerequisites (if any, like setup, operators/versions):

Steps to Reproduce

1. working in "stage"
2. had 1 node with GPUs
3. successfully spawned a notebook using 1 GPU. 
4. closed the notebook
5. removed the GPU node from the cluster (save cost during the night)
6. next day, want to spawn a Notebook
7. GPU node has not been re-added yet. 
8. So display does not propose whether to select GPUs or not. (only container size)
9. However, spawned pod still "rememebers" last settings (gpu: 1) 
10. Pod never runs, because GPU node is not available. 

Actual results:

Pod is pending until either timeout or GPU node is added back into cluster.

Expected results:

Pod should not request GPUs when none are available. 

Reproducibility (Always/Intermittent/Only Once):

Only seen it once, but I'm pretty sure it's always going to happen. 

Build Details:

not sure. 1.6.0 or 1.7.0

Workaround:

• add a GPU node. then select 0 GPU. then remove GPU node.
• 

Additional info: