fix(instances): fix selection loop and require Tailscale for connect

[mfauzaan]

Summary

Fix: instance picker loops forever when there are no main pods

wukong application instances connect built the picker list with a Main: and a Preview: header, then used fixed index math to skip headers:

if instance_name_idx == 0 || instance_name_idx == main_pods.len() + 1 { continue; }

When an application has only preview-labelled pods (e.g. mv-platform Prod / Green), main_pods.len() + 1 == 1, so the only real pod was also treated as a header and the prompt re-appeared on every selection. If it had exited, preview_pods[idx - main_pods.len() - 2] would have underflowed.

Each picker item is now paired with its pod (None for headers), so selection no longer depends on index math. The cursor also defaults to the first selectable pod instead of the header.

Feature: Tailscale confirmation

The Livebook URL is only reachable over Tailscale. Without it, the connectivity test spins for ~100s and then tears the instance down.

• Ask the user to confirm they are on Tailscale before provisioning the Livebook instance. Answering "No" exits without creating any resources.
• Mention Tailscale in the connectivity test spinner and in the failure message.

Test plan

• cargo fmt, cargo clippy (no new warnings)
• Local --features prod build against mv-platform Prod / Green (preview-only pods): picker selects the pod, no loop
• Answer "No" to the Tailscale prompt → exits without provisioning
• Answer "Yes" while on Tailscale → Livebook connects

🤖 Generated with Claude Code

[zacaaron]

🤖