feat(k8s): add agent service and LiteLLM to the Helm chart

[bobbai00]

What changes were proposed in this PR?

The agent-service image is built and runs under single-node compose but had no Helm deployment, and the chart had no in-cluster LLM gateway. This adds both, mirroring the proven preview/production chart while aligning the agent service's env to what the code on main actually reads.

Agent service (gated on agentService.enabled)

• agent-service-deployment.yaml + agent-service-service.yaml, wired to in-cluster service DNS using the env names from agent-service/src/config/env.ts: TEXERA_DASHBOARD_SERVICE_ENDPOINT (webserver), LLM_ENDPOINT (access-control-service), WORKFLOW_COMPILING_SERVICE_ENDPOINT, and a per-CU EXECUTION_ENDPOINT_TEMPLATE.
• A dedicated /api/agents HTTPRoute (REST + the /api/agents/:id/react WebSocket) plus a BackendTrafficPolicy that consistent-hashes on X-Agent-Workflow-Id — agents are held in memory per pod, so a workflow's requests must always reach the same replica (the client already stamps that header).
• Readiness/liveness on /api/healthcheck.

LiteLLM — in-cluster LLM gateway (gated on litellm.enabled)

• litellm-deployment.yaml + litellm-service.yaml + litellm-config.yaml (ConfigMap).
• Postgres persistence on by default: a texera_litellm database is created by the Postgres init script, and the deployment sets DATABASE_URL + STORE_MODEL_IN_DB=true so keys, spend, and model config survive restarts.
• access-control-service wired to LiteLLM (LITELLM_BASE_URL, LITELLM_MASTER_KEY, copilot enabled).

A shared Opaque Secret holds the agent gateway key, the LiteLLM master key, and the provider API keys (supply via --set/override; none committed).

Any related issues, documentation, discussions?

Closes #5269

Also implements the in-cluster LiteLLM Helm support tracked by #4108 (supersedes the approach in #4109).

How was this PR tested?

helm lint and helm template against the chart (subchart dependencies: stripped locally so the render needs no remote charts):

helm lint .                                        # 1 chart(s) linted, 0 chart(s) failed
helm template texera . -f values-development.yaml  # RC=0, 50 objects, no errors

Verified the rendered output: agent deployment env + /api/healthcheck probes; agent-service-svc; the /api/agents -> agent-service-svc HTTPRoute and the BackendTrafficPolicy targeting it (consistent hash on X-Agent-Workflow-Id); LiteLLM DATABASE_URL=postgresql://…/texera_litellm + STORE_MODEL_IN_DB; access-control-service LITELLM_BASE_URL/MASTER_KEY; and the idempotent CREATE DATABASE texera_litellm in the Postgres init script. Not applied to a live cluster.

Was this PR authored or co-authored using generative AI tooling?

Generated-by: Claude Opus 4.8 (1M context)