Refactor Live Client for v3

[davidvonthenen]

This PR introduces the following:

• Refactor the Live client to use IO threading versus async/wait. Found a pretty significant issue which could affect throughput of streaming to the websocket. async/wait is good for non-real-time things as it simplifies things since async/wait uses a single thread for all functions, objects, etc. This turns out to be not so great for IO intensive tasks since pushing the IO will be interrupted from every other thread doing async/wait stuff because they all share a single thread. More info here and here.
• Preserves the async/await Live client implementation to be marked for deprecation in the next major/minor version. A warning will be printed to the user informing them this will be deleted in a future release.
• Introduces a helper class to implement microphone input using pyaudio (backed by portaudio).
• Introduces a simple example "main" style application (located at examples/streaming/microphone) of using the microphone with the Live client.
• Refactors the errors not be global and places them into their own modules.
• KeepAlive messages are disabled by default but can be optionally enabled.

This PR should address the following issues:

• Deepgram Python speech Live. __call__ does not start coroutine for sending #155 - the live client has a newer threading implementation
• Second initialization of LiveTranscription doesn't send data to deepgram #140 - the live client can be stopped and started again now
• LiveTranscription #127 - this should be addressed in the latest implementation

Verified all the examples including the examples/streaming/http using the Legacy client still works correctly.