feat: add experimental global AI assistant option

[elias-ba]

Description

Adds an opt-in checkbox to the AI Assistant UI that routes messages to Apollo's global chat endpoint instead of the separate job_chat/workflow_chat endpoints. The global chat unifies both behind a router that intelligently decides how to handle each request.

Gated behind the existing experimental features preference - only users who have enabled experimental features in their profile see the checkbox.

Closes #4532

Validation steps

1. Enable experimental features in user profile
2. Open a workflow in the editor, open the AI assistant panel
3. Verify "Global assistant (experimental)" checkbox appears next to the PII warning
4. Check the checkbox - badge should change to "Global (experimental)" in amber
5. Send a message - verify it routes through global_chat (check Apollo logs for "HTTP stream connected to global_chat")
6. On the workflow canvas, ask to modify workflow structure - verify "Generated Workflow" card with Apply
7. Navigate to a job, ask to modify code - verify "Generated Job Code" card with Preview/Apply
8. Uncheck the checkbox - verify messages go through normal job_chat/workflow_chat as before

Additional notes for the reviewer

AI Usage

• I have used Claude Code
• I have used another model
• I have not used AI

Pre-submission checklist

• I have performed an AI review of my code (we recommend using /review with Claude Code)
• I have implemented and tested all related authorization policies. (e.g., :owner, :admin, :editor, :viewer)
• I have updated the changelog.
• I have ticked a box in "AI usage" in this PR

[codecov]

Codecov Report

❌ Patch coverage is 98.43750% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 89.58%. Comparing base (7a47ef2) to head (ebc5938).
⚠️ Report is 11 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/lightning_web/channels/ai_assistant_channel.ex	90.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4567      +/-   ##
==========================================
+ Coverage   89.51%   89.58%   +0.06%     
==========================================
  Files         441      441              
  Lines       21205    21265      +60     
==========================================
+ Hits        18982    19050      +68     
+ Misses       2223     2215       -8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[hanna-paasivirta]

@elias-ba When I try a conversational turn with no YAML generation in the workflow chat (no experimental features turned on) I get some errors:



Sometimes the streamed answer disappears and it seems to regenerate a new answer, which in this case was malformed:

I'm not sure why at this stage, but I'll continue testing.

[hanna-paasivirta]

@elias-ba When I try a conversational turn in the global chat (experimental features turned on), I get an error similarly:

[josephjclark]

Hi @elias-ba - haven't run deep testing but the behaviour looks generally right. Happy with the UI.

I'm not seeing any streaming, is that expected? And @hanna-paasivirta do we have many streaming events integrated into the global assistant? It's so slow it's going to be very important we give that feedback to users.

I've seen some very bad responses come back from job chat - just wanted to see how broken it is and yes, it feels a bit broken. I think I'm OK with that - it depends on if the problems are frontend or backend. I might take a look at the implementation this afternoon just to better understand what we have here.

[hanna-paasivirta]

@josephjclark No I haven't implemented streaming in the global agent

[theroinaochieng]

Elias, on the expected user experience this looks good to me. I wasn't able to validate how the data streams as shown on the validation steps though.

[hanna-paasivirta]

I will go ahead and fix the problem in the screenshots separately (OpenFn/apollo#442). So this PR can be merged.

[josephjclark]

So we think that the outstanding issues are in production or purely on the apollo side - so no action to take here.

We are prioritising streaming in the global chat because I think that's important.

[elias-ba]

Thanks so much for taking the time to test this @josephjclark @hanna-paasivirta @theroinaochieng and for helping merge it !

@josephjclark @theroinaochieng what's happening is that streaming actually works when the global assistant routes a question directly to the workflow or job chat agents. But when it goes through the planner (which handles the more involved, multi-step requests), the planner doesn't stream the response text yet, it just sends a couple of status updates like "Thinking..." and then the full answer lands all at once. That's why it feels slow with no feedback. The good news is that everything on the Lightning side is already set up to handle it, so once @hanna-paasivirta adds streaming to the planner on the Apollo side, it'll just work without any changes here (normally).