parsing the command line is too expensive

[Antoine Martin]

The audio subprocess ends up loading picture and video codecs, initializing dozens of loggers, etc.

This was already improved slightly:

• delaying logging: r20248, r20239, r20238, r20237
• removing unused logging code: r20235
• importing constants without side effects like importing other modules: r20246

[Antoine Martin]

Done for video encoders in r20269 + r20270.
We now launch all sound subprocesses with the extra arguments:

'--windows=no', \
'--video-encoders=none', '--csc-modules=none', \
'--video-decoders=none', '--proxy-video-encoders=none'

This avoids loading the "video helper" class, which would load the video and csc codecs.
The list of arguments can be overriden with XPRA_DEFAULT_SOUND_COMMAND_ARGS.

One area where we could still do better is packet encoding and compression: we load all of them (ie: lz4, lzo and zlib - rencode, bencode and yaml) whether we use them or not. Not a big problem since those are all quite small.

@maxmylyn: just a FYI. The audio subprocess should start a bit faster (barely measurable in most cases) and use less memory.

[J. Max Mena]

Noted and closing.