Add Zig core with Python FFI for high-performance tensor operations

[Copilot]

Implements optional high-performance Zig core for tensor operations with Python FFI bindings. Library remains pure Python/JAX by default; Zig provides native performance when available.

Core Implementation

• Zig library (src/zig/core.zig): Matrix operations (multiply, transpose, trace), metric tensors (Minkowski, spherical polar), C ABI exports
• Python FFI (src/autograv/zig_ffi.py): ctypes bindings with NumPy integration, automatic library detection, graceful JAX fallback
• Build system (build.zig): Cross-platform compilation, optimization levels, test suite

Infrastructure

• Build scripts: Shell/PowerShell for Linux/macOS/Windows
• CI/CD (.github/workflows/zig.yml): Multi-platform testing (Ubuntu, macOS, Windows), Zig 0.11.0/0.12.0, cross-compilation validation
• Documentation: API reference, integration guide, development workflow, implementation details

Usage

import numpy as np
from autograv.zig_ffi import spherical_polar_metric_zig, is_zig_available

if is_zig_available():
    coords = np.array([5.0, np.pi/3, np.pi/2])
    metric = spherical_polar_metric_zig(coords)  # Native Zig performance
else:
    # Automatic fallback to existing JAX implementation
    from autograv import spherical_polar_metric
    metric = spherical_polar_metric(coords)

Architecture

Hybrid design: JAX for autodiff/GPU, optional Zig for performance-critical operations. Zig library usable standalone without Python. Foundation for future JAX/XLA custom operations.

Note: Building JAX/XLA with Zig as primary build system is impractical (complex C++ codebase with Bazel). Implementation focuses on Zig-based operations callable from Python, with documented path to XLA custom calls.

Security: CodeQL scan clean (0 alerts), workflow permissions scoped to contents: read.

Files: 15 modified, +2,879 lines (265 Zig, 489 Python, 1,609 docs, 417 build/CI, 99 config).

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• esm.ubuntu.com
  • Triggering command: /usr/lib/apt/methods/https /usr/lib/apt/methods/https (dns block)
• ziglang.org
  • Triggering command: /usr/bin/curl curl -L REDACTED -o /tmp/zig.tar.xz (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

Original prompt

---

This section details on the original issue you should resolve

<issue_title>rust/zig core pending</issue_title>
<issue_description>- maybe zig bindings for jax/xla?

• FFI to python & publish to pypi
• document new library, unique capabilities, git add, git commit, lint, prose lint, format, test (unit and integration), update docs, update CI, update usage instructions
• the idea is to provide a zig core which can be used directly with zig (obviously) and through python via python-zig bindings

we need to figure out how to use Zig as a build system to successfully build jax and/or xla</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes rust/zig core pending #1

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.