Keep SciPy shape-comment overloads compact

Summary:
Put SciPy overload shape declarations between `overload` and `def` so Ruff preserves adjacent overloads, matching scipy-stubs formatting conventions.

Leave NumPy examples, shape contracts, and the comment parser unchanged. Document the chosen placement in the SciPy demo.

Differential Revision: D127646905

Author: stroxler

tensor-shapes/comment-shape-corpus/scipy/README.md

@@ -58,6 +58,8 @@ maximize SciPy coverage.
 ## Syntax observations
 
 - The scalar `n: int  # shaped: N` form is compact.
+- Overload declarations place `# shapes:` between `@overload` and `def`, so Ruff
+  preserves adjacent overloads without inserting a blank line between them.
 - `.0` preserves dtype annotations. Splitting the type lets a shape comment
   sit directly beside `tuple[int, ...]` instead of using a long selector.
 - An `NDArray` alias hides the shape slot, requiring a full `= payload` replacement.

tensor-shapes/comment-shape-corpus/scipy/cholesky.pyi

@@ -42,9 +42,8 @@ def cho_factor(
         np.dtype[np.bool],
     ],
 ]: ...
-
-# shapes: N
 @overload
+# shapes: N
 def cho_solve(
     c_and_lower: tuple[
         np.ndarray[tuple[int, int], np.dtype[np.float64]],  # shaped: .0 = [N, N]
@@ -55,9 +54,8 @@ def cho_solve(
     check_finite: bool = True,
 ) -> np.ndarray[tuple[int], np.dtype[np.float64]]:  # shaped: .0 = [N]
     ...
-
-# shapes: N, K
 @overload
+# shapes: N, K
 def cho_solve(
     c_and_lower: tuple[
         np.ndarray[tuple[int, int], np.dtype[np.float64]],  # shaped: .0 = [N, N]