fix: include all reachable modules in cache generation via BFS over submodule and macro call edges - #9860

TL;DR

Introduced a unified all_crate_modules_for_cache function to replace duplicated module traversal logic in cache generation.

What changed?

A new all_crate_modules_for_cache function was added to cairo-lang-semantic that performs a breadth-first traversal of all modules reachable from a crate root, following both submodule and macro call edges. This replaces the previous pattern of using itertools::chain combined with module_macro_modules calls scattered across generate_crate_def_cache, generate_crate_semantic_cache, and the lowering cache generation. The module_macro_modules import was removed from both the semantic and lowering cache modules in favor of this centralized traversal.

How to test?

Run the existing cache-related tests to verify that crate cache generation still correctly captures all modules, including those nested inside macro call modules.

Why make this change?

The previous approach duplicated the logic for enumerating all relevant modules in multiple places, and relied on module_macro_modules with an inline chain! pattern that was easy to get wrong. Centralizing this traversal into all_crate_modules_for_cache ensures consistent behavior across def, semantic, and lowering cache generation, and makes the intent — following both submodule and macro call edges — explicit and reusable.