Public issues
  https://github.com/mackron/miniaudio/issues/877
  https://github.com/mackron/miniaudio/issues/881

This is from c89atomic which has been fixed upstream.

Public issue https://github.com/mackron/miniaudio/issues/833

fix typos

Public issue https://github.com/mackron/miniaudio/issues/841

Public issue https://github.com/mackron/miniaudio/issues/836.

This adds the following flags:

  * MA_SOUND_FLAG_LOOPING
  * MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_LOOPING

These can be used to initialize sounds and resource managed data
sources to loop by default. This is the recommended way to enable
looping for streams. The `isLooping` config option in
`ma_sound_config` and `ma_resource_manager_data_source_config` has been
deprecated. If you are using those, you should switch to the new flag
or else you'll get compiler errors when upgrading to a future version.

Public issue https://github.com/mackron/miniaudio/issues/811

When processing a node, miniaudio will read into a temporary buffer
before mixing input attachments. This commit removes the per-node heap
allocation and replaces it with a per-graph stack. This should result
in less memory usage at larger scales, but at the expense of slightly
more usage at smaller scales.

The size of the stack can be configured via ma_node_graph_config. If
ma_engine is being used, it can be done via ma_engine_config.

Previously, processing a node would involve a temporary buffer
allocated on the stack. Because this was fixed size, it would result in
processing being sub-divided into chunks in order to avoid overflowing
that buffer. This becomes an issue when a node needs to have a known
processing size. An example might be some kind of effect that requires
processing be in powers of two.

With this commit, the `processingSizeInFrames` variable in
`ma_node_graph_config` can be used to make it so processing always
happens in fixed sized chunks. In this situations, it's recommended you
always call `ma_node_graph_read_pcm_frames()` with a frame count of a
multiple of `processingSizeInFrames`.

The allocation strategy used here is not optimal and will be improved
in future commits. It currently allocates a buffer per-node, but since
the data contained within it is transient in nature, it should be
possible to use a global fixed sized stack that supports allocating a
variable amount of space within the stack buffer.