RFC: Transmission of binary data with JSON

There's already a good writeup on this topic: http://blog.marcinbudny.com/2014/02/sending-binary-data-along-with-rest-api.html

We need to send binary image data, often in excess of 10MB, between the client and server. JSON Base64 increases byte count by 33%, but also has decode/encode overhead.

Key questions for *your* platform/language:

1. How efficient is your JSON implementation and decoding and encoding enormous base64 strings? With Rust<->Rust we could may see 40ms overhead for 4 codings of a 10MB image.
2. Does your HTTP client support sending multipart requests and parsing multipart responses?

Please speak up if you lack both multipart support and a fast JSON codec.

## JSON implementations

... are free to [set limits on the maximum size of strings or the message itself](http://stackoverflow.com/a/30739035).

Vary widely in speed: https://www.reddit.com/r/programming/comments/3pojrz/the_fastest_json_parser_in_the_world/

Are ubiquitous and usually optimized. Benchmarks of huge string parsing and base64 decoding appreciated.

## BSON

BSON has a native binary datatype, but implementations tend to be less optimized than their JSON counterparts. Benchmarks of huge binary blobs appreciated.

## HTTP libraries

Benchmarks of multipart parsing overhead on your platform would be appreciated.

## When it doesn't matter so much

* We can specify URLs instead of sending input blobs, and offload that fetch (and potentially, have some kind of caching).
* We can ask the server to just send the first image result instead of using multipart. We still have to handle failure JSON, though.

## Non-HTTP interactions

* libimageflow users can send pointers to buffers.
* imageflow_tool can read/write to the local filesystem.

No files or tests are named. Start with the linked writeup, then gather the requested benchmarks for JSON/Base64 and multipart handling on the repository's C# platform; done means documenting a selected transport approach and its trade-offs.

Written by the indexing model from the issue text.