Fix: Make StreamStack thread-local to prevent cross-thread interference

Fixes #274

The StreamStack singleton was being shared globally across all threads,
causing thread interference where one thread could see another thread'\''s
active stream. This resulted in crashes when using cvcuda.Stream in
multi-threaded applications.

Solution: Changed the static singleton to thread_local static, ensuring
each thread has its own StreamStack instance.

Author: kgnandanwar

python/mod_cvcuda/nvcv/StreamStack.cpp

@@ -48,7 +48,7 @@ std::shared_ptr<Stream> StreamStack::top()
 
 StreamStack &StreamStack::Instance()
 {
-    static StreamStack stack;
+    thread_local static StreamStack stack;
     return stack;
 }