fix: implement is_exact_type_of to check dtype and shape for PyArray

[asjad2401]

Problem

PyTypeInfo::is_exact_type_of was not overridden for PyArray<T, D>, so the
default PyO3 implementation (a plain Python type object pointer comparison) was
used. This caused cast_exact and downcast_exact to ignore element dtype and
array dimensionality — a float32 array would silently pass a cast_exact::<PyArray2<f64>>() check.

Reported in #549, also related to #547.

Fix

Override is_exact_type_of with the same implementation as is_type_of, which
calls Self::extract internally. This checks:

1. That the object is a NumPy array
2. That the dimensionality matches D
3. That the element dtype matches T

Test

Added cast_exact_checks_dtype_and_shape to tests/array.rs which verifies:

• cast_exact succeeds when both dtype and shape match
• cast_exact fails when dtype does not match
• cast_exact fails when dimensionality does not match