Configurable autoscaling

[nstogner]

• Add configurable scale-down-delay, autoscaling interval & window, and target (Fixes Model expose scale down delay #202)
• Add full scale-up-to-down integration test
• Update helm chart values
• Update docs (add autoscaling docs, remove how-to info from concepts and break into separate sections)
• Remove resource-profile-override fields from Model spec

[samos123]

Looks good except some minor nits and question on behavior.

[samos123]

Alternative:

apiVersion: kubeai.org/v1
kind: Model
metadata:
  name: faster-whisper-medium-en-cpu
spec:
  features: [SpeechToText]
  owner: Systran
  url: hf://Systran/faster-whisper-medium.en
  engine: FasterWhisper
  minReplicas: 0 # defaults to 0 if not set like before
  maxReplicas: 3 # defaults to 3 if not set like before
  concurrentRequests: 100 # defaults to 100 if not set
  ScaleDownDelay: 60s # defaults to 60s if not set
  resourceProfile: cpu:1

Benefits: simple and backwards compatible.

Ideally we keep things backwards compatible also so our existing tutorials and docs don't all have to get updated.

[samos123]

I think I prefer calling it concurrentRequest