Analyze a failed deploy with AI

The AI reads the failed deploy's log, the failing containers' latest output, the names of the source's files, the compose file and the app's variables (never a secret's value), looks further with the read-only API tools when it needs to, and lists what went wrong. A problem's `changes` would fix it: a variable that isn't secret set to a value, a variable renamed (its value moves with it, secret or not), or new scale factors. Nothing is changed: apply them with createVar, updateVar and updateApp, then deploy again. Counts against the caller's 60 AI answers an hour.

curl -X POST "https://example.com/api/workspaces/string/apps/string/deploys/string/analyze"

{  "problems": [    {      "id": "string",      "severity": "error",      "title": "string",      "detail": "string",      "line": -9007199254740991,      "fix": "string",      "changes": [        {          "kind": "setVar",          "key": "string",          "value": "string"        }      ]    }  ]}