Add prepare phase to encode/decode

[tbekas]

This splits computation into two phases:

1. prepare encode/decode - copies data and or parity into buffers
2. encode/decode prepared - runs leoEncode/leoDecode using buffers prepared in phase 1

It helps with threading by avoiding unnecessary data copying. Phase 1. can be done on the main thread while phase 2. can be done on a worker thread.

Name prepare was used to not conflict with init that's used for object initialization.

[dryajov]

It helps with threading by avoiding unnecessary data copying. Phase 1. can be done on the main thread while phase 2. can be done on a worker thread.

With refc you have separate heaps per thread, so if you copy in the main thread and execute in the child thread, then this is most likely the wrong way to do it. This library should be initialized per thread and not shared between threads as it wasn't design for it. Internally, leopard itself is designed to be thread safe, but the wrapper isn't.