Add float array operations across JVM (SIMD), JS, and Native platforms

[Copilot]

Implements Array[Float] extension methods mirroring the existing Array[Double] API across all three platforms, with platform-optimised implementations.

New files

• src/floatarray.scala — FloatArrays object: select, unique, argsort
• src-js-native/floatarray.scala — JsNativeFloatArrays: while-loop implementations shared between JS and Native (clamping, math unary ops with ! in-place variants, element-wise ops, comparisons, outer, logSumExp, productExceptSelf, cumsum, increments, etc.)
• test/src/floatarray.test.scala — 54 cross-platform float tests

Platform implementations

• JVM (src-jvm/arrays.scala): FloatVector.SPECIES_PREFERRED SIMD via jdk.incubator.vector for arithmetic, math ops, reductions, comparisons, and statistics
• Native (src-native/array.scala): BLAS single-precision routines — cblas_saxpy (+=/−=), cblas_sdot (dot), cblas_snrm2 (norm), cblas_sscal (∗=/÷=)
• JS (src-js/array.scala): while-loop implementations for arithmetic, statistics, and linear algebra

Operations covered

All the standard ops from the double API: scalar and array arithmetic (+, -, *, /, in-place variants), math unary ops (exp, log, sqrt, sin, cos, tan, abs, …), dot, norm, mean, variance, std, cumsum, clamp*, argmax/argmin, outer, logSumExp, pearsonCorrelationCoefficient, spearmansRankCorrelation, covariance, elementRanks, fma, **.

import vecxt.all.*

val v1 = Array[Float](1f, 2f, 3f, 4f, 5f)
val v2 = Array[Float](5f, 4f, 3f, 2f, 1f)

v1.dot(v2)                          // Float: 35.0
v1.norm                             // Float: ~7.416
v1.mean                             // Float: 3.0
v1.exp                              // Array[Float]
v1.clamp(2f, 4f)                    // Array[Float](2,2,3,4,4)
v1.pearsonCorrelationCoefficient(v2) // Float: -1.0
v1 * v2                             // Array[Float](5,8,9,8,5)

Also exports FloatArrays.* from all.scala and adds Array[Float] overloads to dimCheck on JS.