ECM.one_curve can give misleading output

[embray]

Right now the ECM interface is broken for me on Cygwin, such that calls to ecm are segfaulting. That's a separate issue though (no ticket yet). What bothers me in this case is that the implementation of ECM.one_curve hides lower-level errors like this, by catching any ValueError from ECM._parse_output and returning a valid output as if no factors were found (when in fact the ecm process itself crashed so yes technically no factors were found...).

Really an error should have been raised much earlier, probably from ECM._run_ecm, but it only checks for errors by checking for output on stderr. It does not check the return code, though it probably should.