Implement sensible timeout for long-running tasks

I had to cancel https://buildkite.com/bazel/bazel-at-head-plus-downstream/builds/2193 after 13 hours (!). The culprit was [rules_nodejs (rbe_ubuntu1604)](https://buildkite.com/bazel/bazel-at-head-plus-downstream/builds/2193#73a307c3-5c1e-43c0-a737-c2a0cd6ed940): The first task timed out after 8 hours, the second one had been running for 5.5 hours when I cancelled it.

Start with the linked Buildkite build and the rules_nodejs (rbe_ubuntu1604) task that ran for hours. Trace the continuous-integration configuration responsible for task timeouts and determine where a sensible limit should apply. Done means long-running tasks stop within the agreed timeout and the affected CI behavior is verified.

Written by the indexing model from the issue text.