SIMD bytecodes

[nrainhart]

• Adds SIMD bytecodes for loading, storing and adding vector registers (float64 and float32)

• Adds SIMD bytecode for subtracting vector registers (float64)

• Adds SIMD primitive to add Float64Arrays.

• Speeds up object initialization in Intel using SIMD instructions

• Disambiguates how longs are read/written from/to memory - 32 vs 64 bits, signed vs unsigned

• Normalizes shape of interpreter stacks in StackVM and CogVM