Consolidate typing for coords with Coords and StrongCoords aliases (fixes #7972)

[aman-coder03]

this PR addresses issue #7972 by consolidating the typing of coords across the PyMC codebase.

Previously, coords appeared with a mix of types such as:

• Sequence
• Sequence | np.ndarray
• dict[str, Any]

this PR introduces explicit and consistent type aliases:

• CoordValue
• Coords
• StrongCoordValue
• StrongCoords

(similar to the existing Shape, Dims, and StrongDims aliases in shape_utils)

Key changes

• standardizes Model.coords to return StrongCoords
• updates ArviZ conversion to return StrongCoords instead of dict[str, Any}
• removes the mixed use of Sequence, Sequence | np.ndarray, and untyped dictionaries
• improves static typing and downstream library compatibility

Notes

while testing this change, a runtime circular import involving:
pymc.model.core
pymc.distributions.shape_utils
from pymc.model import modelcontext
pymc.model (partially initialized)
was encountered.

Fixes #7972

[ricardoV94]

Why this functional change? Doesn't seem necessary or a good idea

[aman-coder03]

i introduced the lazy import only to work around the circular import that surfaced during the typing consolidation (shape_utils -> model -> distributions -> shape_utils). I agree that changing runtime behavior here is not ideal
if you prefer, i can revert this functional change and instead resolve the circular dependency purely by relocating the typing aliases to a more independent module

[ricardoV94]

Put these in a more independent place to avoid the new circular import issues. Coordinates are not really distribution specific anyway

[aman-coder03]

i can move CoordValue, Coords, StrongCoordValue, and StrongCoords to a more independent location and update downstream imports accordingly. Let me know if you have a preferred target module for these.