Pruner fixes

[elijaharita]

• pass background context to long-running pruner goroutine instead of the context parameter in the poll function
• poll the pruner immediately on startup and periodically to resolve an edge case in which the pruner would never prune when out of disk space - before, the pruner would only poll after receiving a blockstore put, but in the event that the drive was completely full already, the put would fail, and the pruner would never end up polling.

[elijaharita]

@hannahhoward the poll function's call chain goes back to the retrieve function, which has a high probability of being canceled. before this change, if the retrieval that initiated the pruner poll operation was canceled, it would cause the prune to be canceled as well because of the shared context, which doesn't make sense since the prune operation shouldn't care what the original caller is doing since it's running independently in a parallel goroutine.