replace base64 with raw data

[YuchengZhou821]

This pull request refactors the audio streaming code in both audio_input_stream.py and audio_rtsp_input_stream.py to improve how audio data is packaged and transmitted. The main change is switching from a JSON-based response containing base64-encoded audio to a binary format that prepends a JSON header to the raw audio chunk. This should improve efficiency and simplify downstream processing.

Audio streaming protocol changes:

• The generator method in both audio_input_stream.py and audio_rtsp_input_stream.py now yields raw bytes instead of a dictionary with base64-encoded audio. The new format consists of a 4-byte length-prefixed JSON header followed by the raw audio chunk. ([[1]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-74140ed30247fd9aaedddedae151eac279bb8026a26f6021b171d09fdbc3bd99L419-R420), [[2]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-6495332b6ea94450241e9a14106cad03cbda46ebc4e10ea720ba8217a29259a8L417-R418))
• The JSON header now includes a millisecond-precision timestamp (timestamp), and is serialized and prefixed to the audio chunk using struct.pack. ([[1]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-74140ed30247fd9aaedddedae151eac279bb8026a26f6021b171d09fdbc3bd99L440-R451), [[2]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-6495332b6ea94450241e9a14106cad03cbda46ebc4e10ea720ba8217a29259a8L438-R449))

Code and interface updates:

• Updated import statements to include struct for binary packing. ([[1]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-74140ed30247fd9aaedddedae151eac279bb8026a26f6021b171d09fdbc3bd99R8), [[2]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-6495332b6ea94450241e9a14106cad03cbda46ebc4e10ea720ba8217a29259a8R5))
• Audio data callbacks now receive the new binary format instead of a JSON string. ([[1]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-74140ed30247fd9aaedddedae151eac279bb8026a26f6021b171d09fdbc3bd99L440-R451), [[2]](https://github.com/OpenMind/OM1-modules/pull/129/files#diff-6495332b6ea94450241e9a14106cad03cbda46ebc4e10ea720ba8217a29259a8L438-R449))

These changes make the audio streaming more efficient and consistent across both input stream modules.