feat(agents): fold edit-with-AI + dock toggle into one split button

[benjackwhite]

Stacked on top of #2742. Puts back the "Edit with AI" affordance that #2742 removed — but de-dupes it structurally rather than dropping it.

Why

#2742 dropped the contextual Edit with AI header CTA to kill the "two near-identical gold buttons" problem: the CTA and the dock toggle both rendered as gold sparkle buttons sitting next to each other. Fair — but that also dropped the one-click hand-off into authoring the view you're currently on. You now have to open the dock and type the intent yourself.

What

AgentBuilderHeaderControls now renders a single split button instead of two competing ones:

• Primary segment — the contextual edit-with-AI action for the current page (New agent / Edit configuration / Explain this session / …), resolved from the dock's page context. Opens the dock and seeds the matching prompt. This is the core action.
• Trailing segment — opens/closes the dock without seeding, with a distinct sidebar icon and an open/hide tooltip that tracks dock state. So you can peek at or dismiss the existing conversation.

One sparkle (the AI identity), one neutral toggle — no more duplicate gold buttons. Mirrors the existing split-button idiom (SkillButtonsMenu): quill Buttons fused with rounded-s/rounded-e + border-s-0 under a shared TooltipProvider.

Views with no obvious action (Scouts) collapse to the lone open/close toggle. Restores agentBuilderActions (headerActionForPage); the header reads page context from the store, so the existing callers stay prop-free (no context prop re-threaded).

Per your note, the Following badge is not carried over into the header for now.

Test plan

• pnpm --filter @posthog/ui typecheck clean; Biome clean.
• Verified live (CDP): the agent-config view shows Edit configuration + an open/close toggle whose icon and tooltip flip with dock state.
• Scouts tab shows the lone toggle (no contextual action).
• Primary segment seeds the right prompt per page (New agent / Explain this session / …).

🤖 Generated with Claude Code

[github-actions]

React Doctor found no issues in the changed files. 🎉

_{Reviewed by React Doctor for commit e4d73f3.}

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
packages/ui/src/features/agent-applications/agent-builder/agentBuilderActions.ts:29-34
The label "Ask about this agent" creates a slightly different mental model than the seed prompt it triggers. Every other label in this switch is either an imperative directive ("Edit configuration", "Review sessions", "Explain this session") or descriptively aligned with the prompt that follows. Here the label implies an open-ended question while the prompt is actually a directive to explain — and the analogous session view already uses "Explain this session" consistently. Aligning the label would make the button's intent clearer and match the established pattern.

```suggestion
    case "agent":
      return {
        label: "Explain this agent",
        prompt: "Explain what this agent does and how it's configured.",
        agentSlug: page.slug,
      };
```

_{Reviews (1): Last reviewed commit: "feat(agents): fold edit-with-AI + dock t..." | Re-trigger Greptile}