osd/ECBackend: optimize remaining read as readop contain multiple objects

osd/ECBackend: optimize remaining read as readop contain multiple objects

in github.com/ceph/ceph/pull/21911
we s/rop.to_read.swap/rop.to_read.erase/ in send_all_remaining_reads(...),
and then introduce a new little issue: if one of to_read objects meet shard error,
we will only change the uncompleted object's to read, but do_read_op will resend the
completed objects shard read, because we did not modify their to_read's need.

Fixes: https://tracker.ceph.com/issues/46876
Signed-off-by: Zengran Zhang <zhangzengran@sangfor.com.cn>