Various improvements to BLS implementation:

* Switch from rayon to crossbeam
* Allow windows to be reused per batch exp
* Allow batchexp to take vector by value
* Allow access to thread-local engine context
* Allow cloning of Engine
* Clean up wNAF abstractions to reduce heap allocation

Cargo.toml

@@ -10,9 +10,10 @@ version = "0.0.1"
 
 [dependencies]
 rand = "0.3.*"
-rayon = "0.6.*"
 byteorder = "1.*"
 serde = "0.9.*"
+crossbeam = "0.2"
+num_cpus = "1.0"
 
 [dev-dependencies.bincode]
 git = "https://github.com/TyOverby/bincode.git"