(though this seem to only happen with debuggers and (?)broken OSes)

[1] system() blocks SIGCHLD, and is probably not intended for multithreading

We cannot assume that blocking calls in other libraries ignore EINTR anyway, 
so the only safe approach is to block in all threads (you can unblock it in
your thread if you really want it) except the signal handling thread.

 * Don't call var_Get every time we have an image to process and register the variables as commands.

Register variables as callback. Don't store module specific variables in p_libvlc_global. They don't belong there.

if we var_Create with DOINHERIT|ISCOMMAND and ISCOMMAND isn't set on the source variable, then inherit anyway and add ISCOMMAND to the type. 

Allow using an object's psz_object_name as id for the libvlc_global commands "tree" and "vars". Example use:
./vlc -I rc --sub-filter marq@test --no-audio ~/media/redefined-nintendo.mpg
vars test
(instead of "vars 376" ... or whatever the integer id is) 

already existed but was only used by modules as far as is know). This
is done in the module_Need function. Needed module names now have this
syntax: '<module>[@<name>]'. If the @<name> part is present, once the
needed module is found it will change p_this->psz_object_name to <name>.
In about 99% of the module_Need calls, p_this is the module's parent
object so this is ok. The remaining calls won't use this syntax so it's
ok i guess :)

* Add new vlc_object_find_name function. It works like vlc_object_find
but uses a string (name) instead of an integer (type) as its second
argument.

* Change the marq, mosaic and logo commands in rc.c. They now take the
target object's name as first argument. Example:

Launch vlc with:
./vlc -I rc --no-audio --sub-filter "marq@test{marquee=Hello}:marq@testouille{marquee=Test}" ~/media/redefined-nintendo.mpg

Then issue the following command to move the second marq:
marq-x testouille 100

(and while testing I fixed #745)

- Use a single pipe rather two pairs

It uses --mmsh-proxy=xxx or http_proxy environment variable (basic
authentication has not been tested).
We may want to reuse --http-proxy from http instead (and so we will have to
move the option out of http.c)