[cstns]

Closes #8619

Merge order: FlowFuse/driver-localfs#363, FlowFuse/driver-docker#254 and FlowFuse/driver-k8s#415 go first and get released, then this one together with bumping the driver versions. On its own this PR is no worse than today, but it only closes the race once the drivers wait for the launch.

Starting the team broker and suspending it right after could leave the schema agent running with no record of it. The suspend landed while start was still launching the agent, so there was no pid to kill yet, and then the record got deleted. Only possible when the two calls overlap within a few ms, so API callers, not the UI.

The agent row now carries the state through the gap. Start creates it as starting, waits for the driver to launch the agent, then sets running. A suspend while it's starting gets a 409 agent_starting instead of trying to stop something that isn't there yet. Suspend marks it suspending before stopping it, and a start in that window (before the row is removed) gets a 409 agent_suspending rather than seeing a "running" agent and doing nothing. If the launch fails the row is removed so it can't get stuck in starting, and a second start while one is already starting is a no-op.

Things worth a look:

• I went with the row state over a lock in the cache driver since the DB is already shared by every replica and nothing new is needed.
• Two concurrent starts can still create two rows, there's no unique index on TeamId. Adding one needs a look at existing data first (a migration would fail on startup if prod has duplicates), so left out.
• Moving out of suspending still relies on the existing 1.5s timer that deletes the row.

Added route tests for the starting/suspending states, the 409s and the failed launch, all failing before the change. Team broker specs pass. Also raced start and suspend against the local app (localfs) at offsets from 2 to 30ms: with this plus the driver change 0 orphans out of 24 tries and the overlapping suspends got 409s. With only the driver change it was 3 orphans out of 24, with only this one 1 out of 18, so both are needed.

[codecov]

Codecov Report

❌ Patch coverage is 93.75000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 79.15%. Comparing base (fe92734) to head (fe230dd).

Files with missing lines	Patch %	Lines
forge/ee/routes/teamBroker/3rdPartyBroker.js	93.75%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8917      +/-   ##
==========================================
+ Coverage   79.07%   79.15%   +0.07%     
==========================================
  Files         476      476              
  Lines       25890    25903      +13     
  Branches     6886     6889       +3     
==========================================
+ Hits        20473    20504      +31     
+ Misses       5417     5399      -18     

Flag	Coverage Δ
backend	79.15% <93.75%> (+0.07%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 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.