Adds `lab task interactive` CLI command that guides users through selecting a compute provider and interactive task template (Jupyter, vLLM, Ollama, etc.), collecting configuration, launching the job, and polling until the service is ready before printing connection info.

- Fix _dismiss_all to use pop_screen exclusively (avoid screen stack corruption)
- Add minutes_requested to resource collection and launch payload
- Validate required env params (reject empty input)
- Use dedicated widget for connection info instead of appending to artifacts
- Add polling retry for connection info in TUI (up to 60s)
- Set selected_provider before set_options to avoid reactive event race

The /server/info route was removed, causing the CLI status command to
return a 404. Switch to /healthz which is always available.

Instructs agents to run CLI tests after modifying code under cli/src/.

The test was reading real credentials from disk instead of simulating
the not-logged-in state, causing it to exit 0 instead of 1.

Consolidate 9 inline current_experiment checks into a single
reusable function in config.py.

…asks-spec

The backend already stores error_msg in job_data when jobs fail, but
neither the CLI nor frontend ever read or displayed it. Now:
- CLI job info shows the error field and renders FAILED in red
- CLI interactive polling shows the error when a job fails mid-wait
- Frontend JobProgress shows error_msg for FAILED jobs
- TUI and CLI interactive forms show help_text hints and use
  placeholder as default values

The CLI was reading `run` from the gallery entry's `command` field,
but GitHub-sourced templates (e.g. Ollama+Gradio) have no inline
command — the run command comes from task.yaml fetched during import.
The UI works because it refreshes the task list after import.

Now the CLI and TUI modal fetch the imported task after import and
read `run`/`setup` from the flat task fields, matching the UI behavior.
Also fixes reading task fields as flat dict instead of nested config.

Add validation in the launch endpoint to reject empty commands with a
clear 400 error. Also improve remote_trap to set job status to crashed
with a descriptive message instead of printing a cryptic usage string.

Match the UI behavior where polling only continues for LAUNCHING,
INTERACTIVE, WAITING, and RUNNING states. Also simplify CLI launch
payload to rely on backend command resolution instead of fetching
the imported task client-side.

When the CLI/TUI doesn't send github_repo_url, github_repo_dir, and
github_repo_branch, fall back to the stored task's fields so the repo
is cloned into the workspace before the run command executes.

Show a spinner with elapsed time while no logs are available, remove
it once real log lines arrive. Filter out "No log files found" server
response so it doesn't display as a log line.

Match the API's transformerlab dependency version with the SDK release.

…ransformerlab/transformerlab-app into add/cli-interactive-tasks-spec