Skaffold dev with HPA setup in the cluster

Originally filed here with all details: https://github.com/GoogleCloudPlatform/cloud-code-vscode/issues/350

I am not sure if this is a bug or an intended behavior, but when I use Cloud Code with Skaffold in my cluster with HPA enabled, I see that new replicas are created every time I do `Run on Kubernetes` even when `targetAverageUtilization` is not reached and takes close to 5-6 minutes to scale down back to 1 whereas this issue does not happen with HPA disabled.

All the screenshots and details showing the same have been attached in the linked issue.

### Information

- Skaffold version: v1.17.2
- Operating system: Ubuntu 20.04
- Contents of skaffold.yaml:

```yaml
apiVersion: skaffold/v2beta10
kind: Config
profiles:
- name: development
build:
artifacts:
- image: asia.gcr.io//
context: .
sync:
infer:
- packages/**/src/**/*.*
- packages/**/schema/**/*.gql
- packages/**/*.md
- packages/**/*.js
local:
push: true
tryImportMissing: false
useBuildkit: true
concurrency: 0
deploy:
helm:
releases:
- name:
chartPath: ./charts/
valuesFiles:
- ./deployment/dev/global/values.yaml
- ./deployment/dev/global/secrets.yaml
artifactOverrides:
image: asia.gcr.io//
namespace: default
skipBuildDependencies: true
useHelmSecrets: true
imageStrategy:
helm: {}
flags:
upgrade:
- --install
statusCheckDeadlineSeconds: 500
```

### Steps to reproduce the behavior

- Setup a HPA in the cluster
- Do `Run on Kubernetes` with cleanUp disabled
- Stop it, and after sometime change something and do `Run on Kubernetes` again
- Replicas are scaled up and down till the max limit setup in HPA