control/media_descriptor.c: Use vlc_gc_* function to prevent input_item_t leak. And prefer p_md over p_media_desc as a variable name. 

control/media_descriptor.c: Don't destroy the input_item, because it might still be needed. We don't have any proper way to do that for now, so we simply leaks it(!).

control/event.c: Make sure the callback list is initially set to NULL. (Likely fix the crash reported by Olivier Aubert).

Author comments:

It makes the following changes:
* Changes to vout_subpictures.c to create some new variables which
renderers are free to use to render time-specific subtitling
information, such as karaoke. One of these variables allows for the text
region to not be permanently converted to YUVA/YUVP - so that more than
one pass can be made on the subtitle - so that it will update correctly
over time. If a rendering module doesn't use or change any of the
variables it behaves the same as usual, with the text region being
rendered just the once to a YUVA/YUVP region and then just blended to
the video on future passes.

* Changes to alignment implementation: no longer use the i_text_align
field of font_style_t at all; require the alignment to be correctly
setup solely in the subpicture_region_t's i_align field (this is so that
the alignment initially inherited from a style can be overridden). This
meant minor changes to freetype.c and quartztext.c as well as the
deletion of the i_text_align field in vlc_osd.h. It also involved some
changes in subsdec.c, where most of the work in this patch occurs.

* Minor change to quartztext.c to correct the interpretation of font
size parameter, making it consistent with a similar change being made in
subsdec.c, and to strip out multiple whitespace in html subtitles
(Similar changes to the freetype.c module have already been made as part
of a separate larger patch submission to fix problems with bidirectional
styled text in that module)

* Adds 2 new fields for carrying karaoke specific information to the
text_style_t struct in vlc_osd.h. Changed the default_text_style's
definition to agree with the changes made in the fields of text_style_t
struct.

* Support for <image> and <karaoke> subtitles in subsdec.c. (Full
working karaoke also requires an as-yet unsubmitted patch to the
renderers to implement)

* Support for more than one subpicture_region_t per timestamp, with each
subpicture region being able to have its own alignment and margins and
type: text or image

* Better calculation of plain-text versions of html subtitles

NB: This patch still makes use of a mechanism of explicitly requesting
the sdl_image module to decode any images it wants. The main reason for
this is that the bmp decoder in ffmpeg (the image decoder with the
highest score) presently only handles 16, 24 and 32 bit depths. An
alternative is for us to switch off the bmp support in ffmpeg for the
time being and the segment of code in subsdec.c that explicitly requests
this module (along with one line added to sdl_image.c) can be removed.
I'm not fussed by either implementation, but this way probably affects
fewer other modules, so have submitted it this way.

JavaScript API to change the selected teletext page from IE ActiveX, Mozilla/Firefox, Safari webbrowser plugins.

Added transparency to vbi pages and added a callback to vbi-opaque setting to change it at runtime from the gui. Threw some compiler fixes and cleanup in the mix.

This patch is a hack to force usage of zvbi teletext decoder. It cannot be resolved cleanly because the codec telx.c and zvbi.c require different offsets in the input buffer, while using the same FOURCC. This is a conflict which should be resolved in another way.

Thanks to sam and pdherbemont (blame them too if this breaks everything!)

Thanks dionoea!