This is the same issue as the earlier commit that fixes the non-UWP
enumeration for WASAPI.

There are two problems with this:

 * The pInfo buffer is being accessed when set to NULL
 * The count is not getting set correctly when pInfo = NULL

This config doesn't really make sense any more because we are now using
dmix/dsnoop for shared mode, and hw for exclusive mode. Format
conversion is now done via mini_al itself which makes plughw pretty
much useless for us at this point.

Having thought about this, it feels like a bad idea to exclude devices
by default because it might get a bit confusing for the end user. The
option to exclude the null device has been swapped around, so now the
config is called "excludeNullDevice" rather than "includeNullDevice".

The buffer size should be set before periods.

Relates to issue #1.

 * By default, device enumeration will now only enumerate over unique
   card/device pairs. Applications can enable verbose device
   enumeration by setting the alsa.useVerboseDeviceInteration context
   config variable.
 * By default, the "null" device is excluded from enumeration. This can
   be changed by setting the alsa.includeNullDevice context config
   variable.

Relates to issue #2.

Possibly related to issue #1.

This adds the mal_context_config structure and changes the
mal_context_init() API to take a pointer to a config object.

It appears snd_pcm_mmap_begin() is returning a frame count of 0 (which
according to the documentation is apparently possible) which mini_al is
not handling properly.

Thanks for that one, Clang!

  * Fix link errors with Clang
  * Add support for VC++ 6. Note that WASAPI and DirectSound are currently
    disabled for this compiler.