[Dhirenderchoudhary]

Fixes #433

Why

The canvas offered Deploy for speech and image nodes with engines that no compute pool can run (speaches, generic). The deploy could not start, and since the canvas only lists READY/RUNNING deployments, the node stayed on "Deploying..." forever.

What changed

• Speech to text and text to speech no longer offer Deploy. When nothing matches, the node says it needs an external deployment and links to New Deployment.
• Image generation now recommends sd-turbo on inferia-diffusion, using the same job spec New Deployment builds (buildDiffusionSpec).
• The canvas polls /deployment/status/{id} for each deploy it started. If the deploy ends FAILED, TERMINATED or STOPPED, the node shows "Deploy failed" with the error and a Retry button.

Scope

• This doesn't add a speaches recipe to the worker. Speech stays external until there is one.
• sd-turbo on inferia-diffusion is not tested on a real GPU pool yet.

Verification

• vitest: the new readiness tests failed before the fix and pass now. One of them checks that every catalog engine is in the worker recipe list. 71/71 workflow tests pass.
• Ran the dashboard against my local stack with Playwright, with deploy and status responses stubbed: speech nodes show the external banner, the image deploy sends inferia-diffusion with its image, and a FAILED status turns into "Deploy failed" plus Retry within one poll.

[Priyanshu-u07]

LGTM
Speech nodes now ask for an external deployment instead of offering a Deploy that can't start, image generation uses inferia-diffusion with the same spec as New Deployment, and a failed deploy shows "Deploy failed" with a working Retry instead of staying on "Deploying…"