feat: retry nodes in downloader

[Frando]

Description

adds retries to #2085:

• after a node fails to dial, or fails during a transfer with IO errors, we move the node into a delay queue, with an increasing timeout. once the delay expires, the node may be dialed again.
• if the head of the download queue can only proceed with nodes which are currently in the retry queue, we park the download in a separate set from the main queue, and proceed with the next download
• once a retrying node succeeds to reconnect, we unpark all parked hashes for which this node is a candidate

Notes & open questions

Has a smoke test for retry behavior. Adding more tests would be great.

Change checklist

• [ ] Self-review.
• [ ] Documentation updates if relevant.
• [ ] Tests if relevant.