fix(active_job): don't clone hub when thread already has one

When a job runs inline inside a Rack request (e.g. ActiveStorage::AnalyzeJob
with perform_enqueued_jobs = true in tests, or inline adapter), the rack
thread already has a hub cloned by CaptureExceptions. Unconditionally
calling clone_hub_to_current_thread on any non-main thread replaced that
hub with a fresh clone of the main hub, wiping out the request transaction.
Downstream instrumentation then called methods on nil and Rails returned 500.

This manifested as a flaky failure on JRuby (where rack threads are real OS
threads, never Thread.main) but not on CRuby (where rack-test runs on the
main thread, so the Thread.current != Thread.main guard fires differently).

Fix: only clone when the thread doesn't already have a hub. This preserves
the lazy-init behaviour for genuine new worker threads (which have no hub
yet) without destroying the hub of threads that were already set up by the
Rack middleware.

Co-Authored-By: github-copilot <noreply@example.com>

Author: solnic

sentry-rails/lib/sentry/rails/active_job.rb

@@ -98,7 +98,14 @@ def record_producer_span(job)
           end
 
           def record(job, trace_headers: nil, user: nil, &block)
-            Sentry.clone_hub_to_current_thread if Thread.current != Thread.main
+            # Only clone the hub onto this thread if it doesn't already have one.
+            # Cloning unconditionally on any non-main thread would destroy the hub
+            # that the Rack middleware set up for the current request, causing the
+            # request transaction to be lost when a job runs inline (e.g. with
+            # perform_enqueued_jobs = true in test or async adapter inline mode).
+            unless Thread.current.thread_variable_get(Sentry::THREAD_LOCAL)
+              Sentry.clone_hub_to_current_thread
+            end
 
             Sentry.with_scope do |scope|
               begin