Buffer Improvements: write new S3 buffer implementation

[tobz]

Part of the Buffer Improvements RFC (RFC, #9476)

We need to write a new buffer implementation that can use S3 as the backing store.

The RFC itself covers a lot more about the goals, overall, with the notion of external buffers and why users want them. We'll cover the high-level items here:

• batching will be required (this is an analogue to the disk buffer flush interval, but also we probably need to support size-based, because high event throughput could lead to memory usage spikes otherwise)
• we have no ordering requirements going in or coming out
• we will follow the S3 sink in terms of design: users will have to set up an SQS queue for S3 notifications in order to let the receiver side of the buffer do its thing

Things without concrete answers at the moment:

• what does the buffer configuration look like? now we have to consider things like auth, batching, etc