use c2-chacha for stable, runtime-dispatched SIMD

Performance:
gen_bytes_chacha20: 254 MB/s -> 603 MB/s [on a Xeon L5630 (SSE4.1)]

Minor version bump:
the only breaking change is that no-std builds now require
default-features=false (std is required by default for runtime cpu
detection; no_std builds will use the best implementation supported by
the target-features/target-cpu enabled at compile time)

New functionality:
ChaChaXRng is parameterized by round count at compile time. Convenient
aliases for the typical 20/12/8 round implementations exposed. ChaChaRng
is aliased to ChaCha20Rng for backward compatibility.

Closes #667

Cargo.toml

@@ -70,7 +70,7 @@ average = "0.9.2"
 # Only for benches:
 rand_xoshiro = { path = "rand_xoshiro", version = "0.1" }
 rand_isaac = { path = "rand_isaac", version = "0.1" }
-rand_chacha = { path = "rand_chacha", version = "0.1" }
+rand_chacha = { path = "rand_chacha", version = "0.2" }
 rand_xorshift = { path = "rand_xorshift", version = "0.1" }
 rand_distr = { path = "rand_distr", version = "0.1" }
 

rand_chacha/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "rand_chacha"
-version = "0.1.1"
-authors = ["The Rand Project Developers", "The Rust Project Developers"]
+version = "0.2.0"
+authors = ["The Rand Project Developers", "The Rust Project Developers", "The CryptoCorrosion Contributors"]
 license = "MIT/Apache-2.0"
 readme = "README.md"
 repository = "https://github.com/rust-random/rand"
@@ -19,7 +19,13 @@ travis-ci = { repository = "rust-random/rand" }
 appveyor = { repository = "rust-random/rand" }
 
 [dependencies]
+generic-array = "0.13"
 rand_core = { path = "../rand_core", version = "0.4" }
+c2-chacha = { version = "0.2", default-features = false }
 
 [build-dependencies]
 autocfg = "0.1"
+
+[features]
+default = ["std"]
+std = ["c2-chacha/std"]