fix: stale slot indices, cascade OOB, pending over-alloc, collect_dependents bounds, size_t format on Windows

- stk_poll: unify load loops to always append on compacted array, removing stale pre-compaction slot indices
- stk_poll: heap-allocate cascade_indices per iteration, removing fixed 256-slot stack bound
- stk_pending_retry: shrink module array to actual count after retry loop completes
- stk_collect_dependents: add capacity parameter and bounds guard before index write
- stk_log_modules: cast module_count to unsigned long for C89 portable %lu on Windows

README.md

@@ -14,7 +14,7 @@ It is designed to run on modern systems running POSIX and Windows using C89.
 - **Hot-swapping** of modules at runtime
 - **Cross-platform** (Linux, BSD, Windows, macOS)
 - **Optional WASM support** for multi-language modules (planned)
-- **Developer tools**: lightweight metadata, logging/tracing, and dependency management (in progress)
+- **Developer tools**: lightweight metadata, logging/tracing, and dependency management
 - **Minimal, portable API**
 
 ---
@@ -234,7 +234,7 @@ stk_init();
 
 ## Project Status
 
-**Current Version:** 1.0.0-pre.7
+**Current Version:** 1.0.0-pre.8
 
 ### What Works
 - Cross-platform module loading and hot-reloading