feat: add usable Agentic Harness goal workflow

[moortekweb-art]

Summary

• add a complete /harness operator surface with a free-form goal composer
• support provider-neutral OpenAI-compatible endpoints and managed Harness routes
• show startup blockers, progress, review actions, evidence, files, and terminal outcomes
• add durable polling and duplicate-submission protection
• isolate provider, managed-Harness, and general Harness proxy credentials
• add beginner-oriented empty, setup, running, blocked, review, and completed states

Why

The previous Harness pages were placeholders: a new user could not clearly enter a goal, understand which model or executor would run it, or follow the task until it was verified or blocked. Proxy boundaries also needed to prevent credentials intended for one backend from being forwarded to another.

This change turns the integration into a usable goal workflow while leaving execution and verification authority with Agentic Harness.

User impact

A user can open Local Studio, choose any supported OpenAI-compatible model or managed Harness route, type a goal, and follow it through completion in one interface. The implementation is not tied to a private Node1 configuration; Node1 vLLM is one compatible deployment.

Checks

• npm run check — passed (format, lint, type checks, unit tests, and production build)
• proxy boundary and credential-isolation tests passed
• Harness page model and beginner-state tests passed
• live end-to-end task completed through the Local Studio surface using an OpenAI-compatible Node1 vLLM route, with exact file and verification evidence
• git diff --check

This is intentionally a draft against dev. It does not merge, release, or alter the upstream deployment.