Errors during project clone are not reported

[amisevsk]

Description

When the project clone init container encounters an issue, it prints the error to logs and exits with a non-zero exit code. This causes it to be restarted by k8s, losing the error log (the second time it runs, the default behavior is to not touch existing directories). This means that

1. Hitting an OOMKilled or similar error means that the repo is left in an inconsistent state (basically, whatever state it was in when the container died -- see project clone operation is altering files after the clone operation #447)
2. Some configuration errors (e.g. a checkoutFrom that cannot be resolved) are suppressed and the workspace starts with the default checkout. A second invocation of the project clone init container wipes out the original log.

Options

1. Catch restarts in the project clone container and fail workspaces that fail to clone projects.

   • This is tricky because it's unclear how to propagate the project-clone failure reason to DevWorkspace status

2. Always return success from project clone to keep logs available -- maybe print errors to a file in $PROJECTS_ROOT to make it more likely that an issue will be seen?