Retry pre-import failures at the manifest level

Problem

While importing large repositories as part of gitlab#350920 (closed), we're especially prone to migration failures due to transient issues. For example, if pre-importing a repository with 50k tags, we may well get up to 40k and then face a transient network timeout that will abort the whole pre-import.

To protect against these events we should handle transient errors at the manifest pre-import level and retry once before giving up.