Speed up parallelized SFTP reads when read size goes beyond end of file

This commit speeds up SFTP read calls when the requested read size is
much larger than the size of the file. Instead of continuing to read
past the end, the parallel read algorithm now stops issuing new read
requests as soon as an earlier read completes with an EOF error. While
calling read() with no length is still the best option if you want to
read a whole file, this optimization may be helpful in some cases.

Thanks go to Maximilian Knespel for reporting this issue and doing
performance measurements on the code before and after the change!

Author: ronf

asyncssh/sftp.py

@@ -670,7 +670,7 @@ async def iter(self) -> AsyncIterator[Tuple[int, _T]]:
                         self._pending.add(asyncio.ensure_future(
                             self._start_task(offset+count, size-count)))
                 except SFTPEOFError:
-                    pass
+                    self._bytes_left = 0
                 except (OSError, SFTPError) as exc:
                     exceptions.append(exc)