Add retry on transient fetch failures in the discovery fanout

[JSONbored]

Problem: A network blip while fetching a repo's policy docs or target issues is silently swallowed into a warnings list, dropping that repo's results for the entire discover run — with no retry at all.

Area: AMS / Miner

Proposal: Add retry with backoff to the fanout's fetch calls, matching the retry discipline already recommended for the CI/gate-verdict pollers above.

Deliverables:

• Retry wrapping the fanout's repo-doc and issue-fetch calls.

Acceptance criteria:

• A simulated transient failure during discover is retried rather than silently dropping that repo.

Resources:

• The opportunity-fanout module.

[linear-code]

JSO-3490