Stabilize NumPy 2.x compatibility with safe dependency ranges

Update dependency constraints to support NumPy 2.0/2.1 while maintaining
backward compatibility with NumPy 1.26.x:

- numpy: >=1.26,<2.2 (avoid experimental 2.2+)
- numcodecs: >=0.16,<0.17 (NumPy 2.x compatible version)
- zarr: >=2.18.3,<3.0.0 (unchanged, NumPy 2.x compatible)
- statsmodels: >=0.14.2 (unchanged, NumPy 2.x compatible)

Removed outdated comment about zarr/numcodecs 0.16.0 incompatibility
as numcodecs 0.16.x is now required for reliable NumPy 2.x support.

This enables testing against NumPy 2.0 and 2.1 in CI while excluding
bleeding-edge 2.2+ versions that may introduce breaking changes.

Author: adilraza99

pyproject.toml

@@ -21,15 +21,13 @@ license = "MIT"
 
 [tool.poetry.dependencies]
 python = ">=3.10,<3.13"
-numpy = "<2.1"
+numpy = ">=1.26,<2.2"
 numba = ">=0.60.0"
 llvmlite = "*"
 scipy = "*"
 pandas = "*"
 zarr = ">=2.18.3, <3.0.0"
-# zarr<3.0.0 is not compatible with numcodecs 0.16.0
-# https://github.com/zarr-developers/zarr-python/issues/2963
-numcodecs = ">=0.13.0, <0.16"
+numcodecs = ">=0.16, <0.17"
 dask = {version="*", extras=["array"]}
 distributed = "*"
 fsspec = "*"