Description of problem:

The new GPU selector field in Model Serving is bound between 0 and max_gpus available if using the `-`/`+` symbols, but the text field allows arbitrarily large numbers to be input. The request for the model server can then be made, which will likely be never satisfied (with no feedback given to the user - unless they try deploying an actual model).

When going to very high numbers, there seems to be an additional bug.
10000000000000000 becomes 10P in the YAML of the model server and the `-` button does not appear to work anymore.
Adding a 9 at the end of 999999999999999 turns it into 10000000000000000.
Adding a 9 at the end of 10000000000000000 turns it into 100000000000000020.
Adding a 9 at the end of 100000000000000020 turns it into 1000000000000000300.
Adding a 9 at the end of 1000000000000000300 turns it into 10000000000000002000.
Adding a 9 at the end of 10000000000000002000 turns it into 100000000000000020000.
Adding a 9 at the end of 100000000000000020000 turns it into 1; `+`/`-` start working again.

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

RHODS 1.27 with GPUs available

Steps to Reproduce

1. Install RHODS
2. Provision GPUs
3. Create a Data Science Project
4. Configure a Model server with an arbitrarily large number of GPUs requested

Actual results:

User can request more GPUs than are available.

User can request impossibly high numbers of GPUs.

Expected results:

User is prevented from requesting a number of GPUs that does not make sense; if requesting more GPUs than are available in the cluster (e.g. for a manual override of GPU discovery) there should be feedback given that the model server cannot be deployed as is.

Reproducibility (Always/Intermittent/Only Once):

Always

Build Details:

Workaround:

Additional info: