It is more reactive and let decoder do proper late/early checks.

This does not work well. It seems that a high level of audio is buffered
inside aout after the input fifo (mixer or output one).

It allows removing partially a ugly input dependency in vout as well as
finer control at decoder side.

It allows removing an ugly input dependency in aout as well as finer control
at decoder side.

It allows replacing --codec dummy --dummy-save-es by --codec dump.

For lack of a better solution, I assume that Windows has a finite
paths length limit to prevent infinite recursion. Otherwise we need
dirfd() and unique inodes.

Forward port: [1a5a6b89692f2e6761b6ecd9ae9ee6e50b6e4138]

(cherry picked from commit b0438df5b9a5a7976be069ec173bb2077f97fd58)

(cherry picked from commit 31a711e42ea80553aca57626903cdd822c8c1bea)

(Not that I expect this to work at run-time)

forwardport [9439810b]

The core currently cannot handle a target name

We now account for this and use an "OpenPanel" instead of a "SavePanel".

Disconnects http-connection before trying with authentication info.
This enables to use http-authentication sources without leaking fds

Kinda work-around