Enhance ArrayUtils.reconstruct with Flat Output Support

[afshawnlotfi]

Overview

This PR enhances the ArrayUtils.reconstruct method in both Python and TypeScript to support returning flat 1-D arrays, providing more flexibility for different use cases like GPU buffer consumption.

Key Changes

Core Feature: flat Parameter

• Added flat: bool = False parameter to ArrayUtils.reconstruct() in both Python and TypeScript
• When flat=True: Returns a single flat TypedArray / ndarray (useful for GPU vertex buffers)
• When flat=False (default): 2-D+ arrays are split into per-column TypedArray[] in TypeScript, or reshaped to original dimensions in Python

TypeScript (typescript/src/array.ts)

• Refactored reconstruct() to reduce code duplication across encoding branches
• Return type changed to TypedArray | TypedArray[]
• 2-D arrays with shape [N, stride] now return an array of stride column arrays by default

Python (python/meshly/array.py)

• Added flat parameter with docstring explaining behavior
• Simplified decode logic by removing inline comments and consolidating reshape step

Mesh Decoding

• Mesh.decode() explicitly passes flat=true to always return flat arrays (mesh data is consumed as interleaved GPU buffers)
• PackableWorkerClient.decodeArray() also uses flat=true

Test Updates

• Added test case for 2-D array column splitting behavior
• Updated Python integration tests to expect TypedArray[] for 2-D fields like inertia_tensor and collision_points

Documentation

• Updated both Python and TypeScript READMEs with usage examples
• Added clarifying note that Mesh.decode always returns flat arrays
• Updated API reference to reflect new return types and flat parameter