Fix block stream retry logic for sequential fetching

[sashaodessa]

retry each block number in NewBlocks before advancing to the next; break out of the fill loop only after exhausting retries for that block

[mattsse]

I think this should just continue here?

[sashaodessa]

I think this should just continue here?

@mattsse We need to break out of the loop once the per-number retries are exhausted. Continuing here would skip over the missing block, leaving a hole at self.next_yield; newer blocks would pile up in the cache but couldn’t be yielded until that hole is filled, so we’d waste RPC calls and still block the stream. The previous implementation also break’d on an unrecoverable failure, so this keeps the same behaviour while making the retry budget actually per block.