trial_build: Auto-retry

It seems like occasionally trial_build fails early for a project because of some network issue. We should have some retry logic to re-do a build (if the failure happens say within 5 mins).

Locate the trial_build entry point and trace how early build failures, especially network failures, are reported. Determine where a retry limited to failures within the first five minutes would fit, then verify that the build is retried only under that condition and that other failures remain unchanged.

Written by the indexing model from the issue text.