ucode: add --skip-preflight to skip per-launch auth/gateway re-validation

[Edwinhe03]

What

Every ucode <agent> launch re-runs configure_shared_state, which does an auth login check, a token fetch, an AI Gateway v2 probe, and (outside provider mode) model discovery — ~5–10s of network round-trips — even when a prior ucode configure already validated all of it. Managed/headless launchers (e.g. omnigent) that run configure once and then launch repeatedly pay this on every turn.

Change

Add a launch-only --skip-preflight flag (distinct from the configure-only --skip-validate, which skips the post-configure model smoke test). It threads through _launch_tool as configure_shared_state(skip_preflight=True):

• configure_shared_state assembles the shared workspace state (profile, base URLs, auth mode) up front, then returns early when skip_preflight is set; otherwise it falls through into the preflight block (auth + AI Gateway validation, then model discovery). One function, one shared assembly — no parallel state builder and no inverted not skip_preflight guard.
• The profile is resolved from ~/.databrickscfg at most once per launch — in the skip branch (trusting the prior configure) or after databricks auth login on the full path, never both.
• The PAT/bearer is already exported by apply_pat_environment, and the AI Gateway was verified by the earlier configure, so the skipped work is redundant. Previously-discovered model lists are preserved (not clobbered with empties).

Wired into every launch command (codex / claude / gemini / opencode / copilot / pi). Default off — behavior is unchanged unless a caller opts in.

--help

$ ucode codex --help

 Usage: ucode codex [OPTIONS]

 Launch Codex via Databricks.

╭─ Options ─────────────────────────────────────────────────────────────────────╮
│ --provider        TEXT  Route through a Unity Catalog Model Provider Service    │
│                         (<catalog>.<schema>.<name>). Skips Databricks model     │
│                         pinning; pass before any `--` separator.                │
│ --skip-preflight        Skip the per-launch Databricks auth + AI Gateway        │
│                         re-validation, trusting a prior `ucode configure`.       │
│ --help                  Show this message and exit.                             │
╰────────────────────────────────────────────────────────────────────────────────╯

Tests

New tests in tests/test_cli.py:

• TestConfigureSharedStateSkipPreflight — configure_shared_state(..., skip_preflight=True) makes no ensure_databricks_auth / ensure_ai_gateway_v2 / get_databricks_token / model-discovery calls (each patched to raise if hit), still rebuilds base_urls + persists state, and resolves the profile when None.
• TestSkipPreflightFlag — --skip-preflight on every launch command threads through _launch_tool to configure_shared_state as skip_preflight=True (and defaults to False without the flag).

$ uv run pytest tests/test_cli.py -k preflight
10 passed, 80 deselected

$ uv run pytest         # full unit suite
858 passed, 8 skipped

ruff + ty clean.

Manual verification

Configured against a real staging workspace (ucode configure --profiles <p> --agents claude --use-pat), then launched the same agent with and without the flag (claude --version execs the real binary and exits, so the launch is observable end-to-end):

$ ucode claude --version                    # DEFAULT — preflight runs
✔ Databricks auth already available for https://…staging.cloud.databricks.com
✔ Unity AI Gateway detected
  Model: system.ai.claude-opus-4-5
✔ Starting Claude Code
2.1.206 (Claude Code)
elapsed: 7.39s

$ ucode claude --skip-preflight --version   # preflight skipped
  Model: system.ai.claude-opus-4-5
✔ Starting Claude Code
2.1.206 (Claude Code)
elapsed: 0.79s

--skip-preflight drops the auth + AI Gateway round-trips (7.39s → 0.79s, ~7s saved), and the agent still launches on the correctly-resolved model — confirming the trusted state (base URLs rebuilt, model lists preserved) is intact.

This pull request and its description were written by Isaac.

[lilly-luo]

this is great! can you put the --help outputs on the PR description and have some examples to confirm this is working? before merge

[lilly-luo]

ideally also rename to --skip-preflight