Remove `pydantic` as dependency for the k8s_service_info library to reduce charm pack times?

### Context

[this comment](https://github.com/canonical/kfp-operators/pull/436#discussion_r1569349087) documents how using the k8s_service_info library increase charm packing time by ~8 min. This isn't a complete blocker, but that's a significant increase and makes running integration tests (local and CI) a lot slower. Should we consider doing something about this?

Options:
* remove the dependency (it is only lightly used, and can be replaced by a dataclass)
* ? Maybe enable caching pip builds between our CI runs so we dont rebuild the pydantic whl every time?

### What needs to get done

See below

### Definition of Done

1. implement something that mitigates the increase in charm packing time caused by the k8s_service_info library