Float32 scalars, arrays and Tensor bridges

import aner.nn

// Literal inference remains Int64 / Float64.
let default_integer = 42
let default_float = 0.5
let exact: Float32 = 0.5
let rounded = Float32(0.1)
let wider: Float64 = rounded
print(type(default_integer)) // Int64
print(type(default_float)) // Float64
print(type(exact)) // Float32
print(type(wider)) // Float64
print(Float64(rounded) == 0.1) // false
print(type(Float32(1.0) + Float32(2.0))) // Float32
print(type(Float32(1.0) + 2.0)) // Float64
print(type(Float32(5.0) / Float32(2.0))) // Float64

var numeric: Number = rounded
print(type(numeric)) // Float32
numeric = UInt8(12)
print(type(numeric)) // UInt8
print(Int16(Float32(3.9))) // 3

let samples = Array<Float32>(size: 3, fill: 0.5)
samples[1] = Float32(0.1)
print(dtype(samples)) // Float32
print(type(samples.sum())) // Float32
print(type(samples.mean())) // Float64
let precise = Array<Float64>(size: 3, fill: 0.1)
let compact = precise.astype("Float32")
let widened = compact.astype("Float64")
print(dtype(compact)) // Float32
print(dtype(widened)) // Float64
print(dtype(compact.to_float())) // Float64

let matrix = NDArray<Float32>(shape: List(2, 2), fill: 1.5)
let packed_tensor = matrix.to_tensor()
print(packed_tensor.dtype()) // Float32
let flat = Array.from_tensor<Float32>(packed_tensor)
let growable = DynamicArray.from_tensor<Float32>(packed_tensor)
let shaped = NDArray.from_tensor<Float32>(packed_tensor)
print(dtype(flat)) // Float32
print(growable.len()) // 4
print(shaped.shape()) // List(2, 2)
let wide = Array.from_tensor<Float64>(packed_tensor.astype("Float64"))
print(dtype(wide)) // Float64
flat.set(0, Float32(9.0))
print(packed_tensor.value(0, 0)) // 1.5: reverse bridge owns independent storage

// Checked failures, intentionally disabled:
// let inexact: Float32 = 0.1 // R1003: implicit precision loss
// samples.set(0, 0.1) // R1003: explicit Float32(...) rounding required
// let overflow = Array<Float32>(2, Float32(3e38)).scale(Float32(2.0)) // R3001: nonfinite result
// let mismatched = Array.from_tensor<Float64>(packed_tensor) // R3001: exact dtype match required