test(stovepipe): add build-slow marker to the fake build runner

[behinddwalls]

Summary

Why?

The fake build runner never reports a non-terminal status — Status decides purely from the build id and always returns succeeded, failed, or an error. That means no integration or e2e stack can reach buildsignal's reschedule branch, so the entire poll loop is exercised only by unit tests with mocked publishers. A defect in the reschedule path is invisible above the unit layer.

What?

Adds a build-slow marker. Trigger encodes a ready-at wall-clock instant into the build id (fake-build-slow-<readyAtMs>-<suffix>) and Status reports running until that instant passes, then succeeded. Encoding the deadline in the id preserves the fake's stateless, decide-purely-from-the-id property, so Trigger and Status can still live in different processes. SlowBuildDurationMs is a package var so tests can tune it.

Also widens marker() to end a token at / as well as & and #. The fake SourceControl builds head URIs as git://<queue>/HEAD, so a marker injected through a queue name sits mid-URI and would otherwise be read as build-slow/HEAD. Existing markers sit at the end of a URI and are unaffected.

No pipeline behaviour changes; this only makes a previously unreachable path reachable in tests.

Test Plan

✅ bazel test //stovepipe/... — covers build-slow reporting running then succeeded, the no-deadline fallback, and marker() against trailing path segments, query separators, and end-of-URI.

Stack

1. @ test(stovepipe): add build-slow marker to the fake build runner #464
2. fix(stovepipe): mint a distinct message id for each buildsignal re-poll #465
3. refactor(stovepipe): replace recorded greenness states with build outcomes #466
4. feat(stovepipe): record the build outcome on the request and free its slot #467
5. feat(stovepipe)!: key the record stage on the request id #468
6. fix(stovepipe): stop wrapping the buildsignal re-poll publish as retryable #469

[mnoah1]

An exported var just to be overridden by tests is a bit of an anti-pattern. We control the runner implementation below, so why not just set a default value New() and let tests set the struct field manually when needed.