[ppraneth]

Closes #9

• Tensors, reductions and printing are generic on the value type T, which defaults to f32, so existing code works unchanged
• LeftComposition can change the value type, e.g. a comparison turns f32 into bool
• Reductions are defined for f32, f64 and i64
• where_ takes a bool condition
• square, negate, scale, clamp and the functions that only move indices are generic
• pow is removed
• Kept on f32 on purpose:
  • zeros, ones and eye have no input to infer the type from