fix(flask): wrap wsgi_app call in try/except to prevent active_requests gauge leak

If wsgi_app() raises an uncaught exception, the active_requests_counter
decrement at the end of _wrapped_app was never reached, causing the gauge
to permanently read high. Kubernetes HPA and similar systems would see
phantom load.

Add a bare try/except that decrements the counter and re-raises on
exception, matching the pattern already used in the WSGI instrumentation.

Fixes #4431

Signed-off-by: alliasgher <alliasgher123@gmail.com>

Author: alliasgher

instrumentation/opentelemetry-instrumentation-flask/src/opentelemetry/instrumentation/flask/__init__.py

@@ -415,7 +415,11 @@ def _start_response(status, response_headers, *args, **kwargs):
                     response_hook(span, status, response_headers)
             return start_response(status, response_headers, *args, **kwargs)
 
-        result = wsgi_app(wrapped_app_environ, _start_response)
+        try:
+            result = wsgi_app(wrapped_app_environ, _start_response)
+        except Exception:
+            active_requests_counter.add(-1, active_requests_count_attrs)
+            raise
 
         # Note: Streaming response context cleanup is now handled in the Flask teardown function
         # (_wrapped_teardown_request) to ensure proper cleanup following Logfire's recommendations