Retry scripts on transient gRPC and pruned-state errors

[janezpodhostnik]

Description

Makes the script runner resilient to failures seen during long-running scans:

• Retry on codes.Unavailable gRPC errors (connection drops); gRPC reconnects on the next attempt, so these are transient.
• When a batch's reference block state is pruned on the execution nodes, resubmit the whole batch at the latest scanned height once it advances past the batch's height, instead of retrying or splitting at a height that can never succeed. If the batch is within PrunedStateFatalHeightGap of the latest scanned height (or nothing has been scanned yet), the error stays fatal.
• All other resubmissions (retry, split, exclude) are also moved to the latest scanned height when it is higher, via AddressBatch.WithBlockHeight, which shares done-tracking with the original batch.
• Reduce the script result channel buffer from 10000 to 500 to apply backpressure to producers earlier and bound memory usage.
• Skip the empty address in event candidate scanning: it is not a real account (e.g. TokensWithdrawn during the initial FLOW mint has no source) and can never be scanned.

Most critical to review: scanner/script_runner.go (new retry-at-latest-height path and fatal-gap check) and scanner/address_batch.go (WithBlockHeight).

---

For contributor use:

• Targeted PR against main branch
• Linked to Github issue with discussion and accepted design OR link to spec that describes this work.
• Code follows the standards mentioned here.
• Updated relevant documentation
• Re-reviewed Files changed in the Github PR explorer
• Added appropriate labels