Yet another set of I/O problems

• pw.x wasn't behaving as advertised. By default, it was still saving the wavefunctions in internal ("davcio") format to file at the end of a run. Not a big deal maybe but it doesn't make sense and wastes I/O and disk.
• This was hiding a bug in one of my recent fixes: the information on the presence of wavefunctions in "collected" format was lost after being read and was no longer available when it was used I hope this will fix at least some of yesterday's problems. More serious and badly needed cleanup postponed to after the release.