* modules/audio_filter/resampler/bandlimited.c: re-enabled the bandlimited resampler. Commit 12102 is actually bogus since the problem is in fact created by the audio output pipeline introducing more latency than what can be compensated by the default input caching.

  Sigmund says that YUVP rendering sucks and that he wants his 256 YUVA renderer back.
  Who exactly remembers why this one was removed ???

* The PTS/PCR repeats we test for should at least continue for 1s before we consider it to be the EOF of a QTSS VoD session.

* if a PTS keeps repeating for a while we assume EOF. This fixes QTSS VoD which VLC kept open after it was finished.
  The limit i chose in Demux() is 5 loops of Demux() with the same pcr. It's based on nothing, but I assumed it was more safe then checking for a single repeat of a PTS value. That might be allowed in some cases????

   and the other way arround. Shouldn't change anything about how the
   thing is run)

* Now uses the fullscreen variable instead of setting i_changes |= FULLSCREEN_CHANGE since this is not handled by vout filters, and variables are.

* Only call gnome_vfs_escape_path_string on the path part of the uri, since the functions in gnomevfs that are supposed to make the host/user/path separation are broken (they will escape passwords too)
* Trying to open a file with http on a system with non latin1 encoding may still fail (testing welcomed).
* Don't call gnome_vfs_make_uri_from_input_with_dirs since this will fail on a path with special caracters, but gnome_vfs_expand_initial_tilde and gnome_vfs_make_uri_from_shell_arg