Python: SnakeCaseNamespaceNames never executes — CrawlTree inside empty-name guard

Nobody has claimed this yet.

## Description

The `PythonRefiner.SnakeCaseNamespaceNames` method (which was removed in #7714 to avoid a breaking change) was intended to convert namespace segments to snake_case at the CodeDOM level before code generation. However, it never actually executed because `CrawlTree` was placed inside the `if (currentElement is CodeNamespace codeNamespace && !string.IsNullOrEmpty(codeNamespace.Name))` guard.

Since `RefineAsync` passes the root namespace (which has an empty name), the guard fails at the root and traversal never starts.

### Impact

- CodeNamespace.Name values remain in their original casing (e.g., `exportNamespace` instead of `export_namespace`)
- The path segmenter and import manager apply `.ToSnakeCase()` downstream, so generated **file paths and import statements are already correct**
- The only observable effect is in the public API export surface, which reflects the raw CodeNamespace.Name

### Why this wasn't fixed in #7714

Simply moving `CrawlTree` outside the guard (the obvious fix) causes a **breaking change**: `GetDifferential()` in `CodeNamespace.cs` strips the namespace prefix by character count (`Name[prefixLength..]`). The `clientNamespaceName` passed to the writer comes directly from the user's config and is NOT snake_cased. After snake_casing CodeNamespace.Name, the prefix length no longer matches, corrupting import path computation.

### Proper fix

To properly fix this, both changes are needed together:
1. Move `CrawlTree` outside the guard in `SnakeCaseNamespaceNames`
2. Snake_case the `clientNamespaceName` before passing it to `PythonWriter`/`PythonRelativeImportManager`

This ensures prefix stripping in `GetDifferential` works correctly with the now-snake_cased namespace names.

### Related

Split out from #7714 to avoid scope creep and breaking changes.

Start with PythonRefiner.SnakeCaseNamespaceNames and RefineAsync, then inspect CodeNamespace.GetDifferential() to understand its prefix slicing. Trace how clientNamespaceName reaches PythonWriter and PythonRelativeImportManager. Done means namespace names are snake_cased without corrupting generated import paths or file paths, with the public API export surface reflecting the converted names.

Written by the indexing model from the issue text.