feat: implement GRPC xDS on merlin standard transformer

[anantadwi13]

Description

This MR is used to introduce xDS GRPC client load balancer to Merlin Standard Transformer.

Previously, the transformer needs to call istio gateway to make sure all requests are balanced across caraml-serving-* pods. But now we support proxyless GRPC call via xDS to reduce gateway costs while still maintaining balanced request distribution.

More about xDS can be seen here https://istio.io/latest/blog/2021/proxyless-grpc/

To enable xDS, you need to configure your merlin config below

...

StandardTransformerConfig:
  DefaultLabels:
    sidecar.istio.io/inject: "true"
  DefaultAnnotations:
    inject.istio.io/templates: grpc-agent
    proxy.istio.io/config: '{"holdApplicationUntilProxyStarts": true }'
  ImageName: ghcr.io/caraml-dev/merlin-transformer:x.x.x # replace x.x.x with version that supports xDS
  FeastRedisConfig:
    ...
    ServingURL: xds:///caraml-serving-redis.caraml-store.svc.cluster.local:50051 # make sure to use xds:///
  FeastBigtableConfig:
    ...
    ServingURL: xds:///caraml-serving-hbase.caraml-store.svc.cluster.local:50051 # make sure to use xds:///

...

Once you applied above config, when you deploy/redeploy a model, you will see the InferenceService resource looks like below

apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: test-pyfunc-model-1-r2
  namespace: caraml-testing-project
  ...
spec:
  predictor:
    containers:
      ...
    ...
  transformer:
    labels:
      sidecar.istio.io/inject: "true"
    annotations:
      inject.istio.io/templates: grpc-agent
      proxy.istio.io/config: '{"holdApplicationUntilProxyStarts": true}'
    containers:
    - env:
      - name: FEAST_STORAGE_CONFIGS
        value: '{ "1": { "redisCluster": { "feastServingUrl": "xds:///caraml-serving-redis.caraml-store.svc.cluster.local:50051" } }, "2": { "bigtable": { "feastServingUrl": "xds:///caraml-serving-hbase.caraml-store.svc.cluster.local:50051" } } }'
      ...
    ...

Modifications

• introduce DNS & xDS client loadbalancing on the merlin standard transformer
• introduce DefaultLabels & DefaultAnnotations to merlin config StandardTransformerConfig. these fields will be used in the InferenceService creation.

Tests

Checklist

• Added PR label
• Added unit test, integration, and/or e2e tests
• Tested locally
• Updated documentation
• Update Swagger spec if the PR introduce API changes
• Regenerated Golang and Python client if the PR introduces API changes

Release Notes

NONE