Implement retry in readonly open

[Ngoguey42]

Since #2119 we are able to detect if a control file is being read with a data race. These data races can be solved by sleeping and retrying. We are however not capable of distinguishing between data races and corrupted control files, this implies that the number of retry should be bounded to fallback on a Corrupted_control_file error.

Also, since the GC finalisation deletes old files, a data race is possible if a RO instance stalls between the read the control file and the opening of the files. The data race would materialise as a No_such_file_or_directory error.

Both these data races could be fixed at once by implementing a retry in the file manager.