PR: Implement support for *Python Array API Standard*.

[KelSolaar]

Summary

This PR implements support for the Python Array API Standard, enabling computations to dispatch onto alternative array backends:

• NumPy (default)
• JAX
• PyTorch (including Apple MPS)

Dispatch is currently opt-in and NumPy-only behaviour is unchanged by default. Once enabled, the backend is selected from the type of the input array. It can be enabled three ways:

1. Environment variable, set before importing Colour:

import os

os.environ["COLOUR_SCIENCE__ARRAY_API"] = "1"

import colour
import jax.numpy as jnp

colour.XYZ_to_sRGB(jnp.array([0.20654008, 0.12197225, 0.05136952]))
# Array([0.7057394 , 0.19248262, 0.2235417 ], dtype=float32)

2. Programmatically, toggle the global state at runtime:

import colour
import torch
from colour.utilities import set_array_api_enabled

set_array_api_enabled(True)

colour.XYZ_to_sRGB(torch.tensor([0.20654008, 0.12197225, 0.05136952]))
# tensor([0.7057, 0.1925, 0.2235], dtype=torch.float64)

3. Scoped context manager (also usable as a decorator), enable for a block only:

import colour
import torch
from colour.utilities import array_api_enable

with array_api_enable(True):
    colour.XYZ_to_sRGB(
        torch.tensor([0.20654008, 0.12197225, 0.05136952], device="mps")
    )
# tensor([0.7057, 0.1925, 0.2235], device='mps:0')

What's added

• Namespace-aware boundary helpers + a full xp_* operation surface in colour.utilities:
  • Namespace resolution: array_namespace, is_numpy_namespace, is_non_ndarray, trace_array_namespace
  • Boundary conversion: as_ndarray, cast_non_ndarray, xp_as_array / xp_as_float_array / xp_as_int_array, xp_astype, xp_ascontiguousarray
  • Shape & manipulation: xp_reshape, xp_squeeze, xp_atleast_1d / xp_atleast_2d, xp_broadcast_to, xp_matrix_transpose, xp_resize, xp_pad, xp_insert
  • Reductions & statistics: xp_average, xp_median, xp_nanmean, xp_trapezoid, xp_gradient
  • Element-wise math: xp_degrees / xp_radians, xp_sinc, xp_round, xp_nan_to_num
  • Linear algebra: xp_lstsq, xp_eig / xp_eigh, xp_create_diagonal
  • Sampling, interpolation & set operations: xp_linspace, xp_interp, xp_select, xp_isin, xp_setxor1d, xp_unique
  • Comparison & testing: xp_isclose, xp_assert_close, xp_assert_equal
• contextvars-backed global state (Array API enablement, domain-range scale, ndarray copy, caching) for thread/async safety.
• SciPy-free, dispatchable kernels replacing solver/interpolator hotspots: correlated colour temperature Gauss-Newton (colour.temperature.common), Jakob and Hanika (2019) trilinear interpolation, etc.
• Default complex precision: COLOUR_SCIENCE__DEFAULT_COMPLEX_DTYPE / set_default_complex_dtype.
• New public API: CIE_illuminant_D_series, msds_CIE_illuminant_D_series, msds_blackbody, msds_rayleigh_jeans.
• Cross-backend testing: an xp pytest fixture parametrising numpy/jax/torch/torch-mps, with mps_tolerance_absolute and mps_xfail markers for float32 precision, plus a cross-backend benchmark suite (utilities/benchmark.py).
• Documentation: a dedicated Array API Support section in advanced.rst.

Performance

Per-suite speed-up vs NumPy (best-of-3, HD inputs): speed-up = NumPy ÷ backend over cases succeeding on both, so higher = faster (e.g. 3.0× = 3× faster than NumPy; < 1.0× = slower). numpy (ms) is the summed best-of-3 over the suite's cases.

Suite	cases	numpy (ms)	jax	torch-cpu	torch-mps
conversion_graph	207	31455.1	4.4×	3.5×	11×
conversion_graph_iterative	4	36874.8	3.3×	1.8×	12×
difference	17	2139.4	3.1×	3.0×	18×
integration_array	2	201.5	644×	1.00×	15×
integration_object	6	3.2	1.3×	0.77×	0.76×
transfer_function	114	3721.2	15×	2.3×	10×
adaptation	7	1066.4	4.4×	4.8×	6.0×
characterisation	3	249.9	6.4×	4.0×	9.5×
recovery_array	4	1878.3	2.7×	3.0×	7.2×
recovery_object	3	452.6	0.65×	0.66×	0.48×
quality_array	4	488.9	2.4×	2.8×	1.5×
quality_object	5	5.1	0.10×	0.39×	0.05×
volume	2	39.5	1.1×	0.98×	1.2×
volume_iterative	2	1856.4	1.00×	0.97×	16×
phenomena	5	172.8	3.0×	4.5×	27×
temperature_array	4	294.7	4.7×	1.6×	18×
temperature_iterative	4	1015.5	0.75×	1.3×	0.53×
blindness	3	163.4	271×	15×	8.1×
contrast	1	78.6	5.1×	3.5×	26×
generators_array	3	51.8	3.3×	1.7×	1.9×
generators_object	6	12.7	1.3×	1.3×	0.84×
photometry	3	0.1	0.08×	0.18×	0.02×
overall	409	82221.8	3.3×	2.2×	8.5×

NumPy is the baseline (1.00×). Measured on an Apple M1 Max (10-core, 32 GB), macOS 15.7, Python 3.13, NumPy 2.3, PyTorch 2.9, JAX 0.8; 409 cases across 22 suites.

Preflight

Code Style and Quality

• Unit tests have been implemented and passed.
• Pyright static checking has been run and passed.
• Pre-commit hooks have been run and passed.
• [N/A] New transformations have been added to the Automatic Colour Conversion Graph.
• New transformations have been exported to the relevant namespaces, e.g. colour, colour.models.

Documentation

• New features are documented along with examples if relevant.
• The documentation is Sphinx and numpydoc compliant.