external storage S3 backend reader can't retry after meet many times of `read: connection reset by peer` error

[lance6716]

Enhancement

Seems the old retry mechanism is for a reader, it can only retry 3 times. When the file reading range is large can each Read is relatively small, it's hard to use 3 retry to finish reading.

I think it's better to tolerate 3 retry inside the Read call

[lance6716]

This causes global sort whole subtask failed and retry from the beginning of subtask