(Bug) Fix InferenceService readiness bootstrap from missing workload conditions

[YouNeedCryDear]

What this PR does

• converts missing workload readiness conditions into explicit Unknown conditions instead of silent no-ops
• seeds top-level and component InferenceService conditions on first observation so fresh raw services do not publish readiness before authoritative workload status exists
• updates readiness-related tests to cover the missing-condition behavior

Why we need it

Fresh raw InferenceService objects can report Ready before DeploymentAvailable / LeaderWorkerSetAvailable has been populated. This closes the bootstrap hole and makes missing workload status block readiness instead of leaking through as Ready=True.

Fixes #545

How to test

• go test ./pkg/controller/v1beta1/inferenceservice/status
• go test ./pkg/controller/v1beta1/inferenceservice/reconcilers/deployment
• go test ./pkg/controller/v1beta1/inferenceservice/reconcilers/lws
• go test ./pkg/controller/v1beta1/inferenceservice/... currently fails in the envtest-backed controller suite in this sandbox with listen tcp 127.0.0.1:0: bind: operation not permitted

Checklist

• Tests added/updated (if applicable)
• Docs updated (if applicable)
• make test passes locally

[gemini-code-assist]

Code Review

This pull request updates the ome-manager and model-agent images to v0.1.5 and refactors status condition handling to use explicit Unknown states during initialization and retrieval. It also adds logic to re-fetch Deployment and LeaderWorkerSet resources to capture status updates. The review feedback points out that context.TODO() should be replaced with proper context propagation and warns that immediate re-fetching might return cached data rather than the updated status from the API server.