Fastly custom runners send no response when the managed logger fails to install #505

Summary

On Fastly, lifecycle::run_custom and lifecycle::serve_custom install the managed logger before the custom callback. When that install fails, they return ManagedLoggerError to the caller without sending a response, so the client's request goes unanswered, health checks included.

Behavior today

At 10166628 (the same files as PR #275's 5c32ae61):

• run_custom runs initialize_for_app::<A>()? before the callback (crates/edgezero-adapter-fastly/src/lifecycle.rs:198), and serve_custom does the same per request (:170). On failure the callback never runs and nothing is sent. Under serve_custom, DeliveryResult::send returns the error without sending (:31-40), and the loop stops.
• The failure is kept for the instance (edgezero-core/src/logging.rs, ManagedLoggerInstallation). The edgezero_logging override is read once per instance, and an unexpected host failure there returns ConfigurationUnavailable (docs/guide/adapters/fastly.md, "Runtime Level Overrides"). So one failed read fails every later request in a reused instance.
• Under Viceroy 0.21.0 the client gets HTTP/1.1 200 OK with content-length: 0. What hosted Fastly answers was not checked.
• The other adapters hand the host an error rather than leaving the request unanswered:
  • Cloudflare run_app maps the failure to WorkerError::RustError("managed logger initialization failed").
  • Spin run_app returns it with ?.
• The Fastly guide does not say what the client receives in this case.

Reproduction

Trusted Server PR IABTechLab/trusted-server#1506, which calls run_custom::<TrustedServerApp, _, _, _> and serve_custom::<TrustedServerApp, _, _, _>:

1. Set [adapters.fastly.logging] endpoint = "ts:log" in edgezero.toml.
2. Build the release wasm and serve it with Viceroy 0.21.0.
3. curl -si /health and curl -si /.

Both answer HTTP/1.1 200 OK, content-length: 0, with no body. Viceroy's stderr shows Trusted Server's report of the returned error: managed logger installation failed: managed logging configuration is invalid. With endpoint = "tslog", /health answers 200 with body ok.

Acceptance criteria

• When managed logging fails before the callback, the client receives an error response (5xx) instead of none, under both run_custom and every request serve_custom stops on.
• The error still returns to the caller.
• The Fastly guide states what the client receives.
• A Viceroy test pins the status.

Related

• Found in the round-2 review of Cut Trusted Server over to EdgeZero PR 275 trusted-server#1506, which pins rc/identity at 10166628 ((DO NOT MERGE) Identity integration branch #495). Trusted Server records the behavior as an open EdgeZero gap and adds no workaround.
• Related: Reject Fastly logging endpoint names the logger cannot install when the manifest loads #506, the endpoint names that fail this way. Part of Platform capabilities for server-side identity on every adapter #468.