pkgs/id tests: bound the unbounded network awaits (live-iroh connects/receives) so a wedged test cannot hang a nix check phase #5218

Surfaced by the code#5195 investigation (2026-11-03, evidence on that issues closing comment). #5195s test proved fully bounded (all waits are clamped tokio sleeps / timeout-wrapped charges), so it cannot explain the 2026-11-03 fleet-switch hang: a workspace cargo test binary sat single-threaded at 0 CPU for 52 minutes inside a nix check phase and had to be killed by hand (the laptop nix config sets max-silent-time = 0 / timeout = 0, so nothing reaps a hung build).

Lead

The id lib test binary carries live-iroh tests whose connects/receives have no outer timeout. An unbounded network await is the only identified mechanism in the id test surface consistent with an indefinite 0-CPU wait. This is a lead, not an attribution — the hang could not be attributed from outside the nix sandbox.

Ask

Audit the id test suites (lib + integration) for unbounded awaits — iroh connects/receives and any other network or channel waits without deadlines — and bound every one with a loud timeout, so a wedged test fails in seconds with its name attached instead of hanging a build forever. If the audit finds the surface already bounded elsewhere, record that; the remaining value is the inventory.

Related

code#5195 (the flake fix; closed), code#5216 (host-side clap stack overflow, separate).

Acceptance criteria

• Every iroh connect/receive await — and any other network or channel wait without a deadline — in the pkgs/id lib and integration test suites is wrapped in a bounded timeout that fails the test loudly with its name attached.
• A deliberately wedged await demonstrates the bound: the test fails within its timeout instead of hanging (evidence recorded on the implementing PR).
• The full pkgs/id test suite passes in the nix check phase / CI with no test deleted or ignored to achieve the bound.