fix: guard TelemetryEventBuffer against re-entrant mutex acquisition

sentry-ruby/lib/sentry/telemetry_event_buffer.rb

@@ -50,6 +50,9 @@ def flush
     alias_method :run, :flush
 
     def add_item(item)
+      # Prevent ThreadError from re-entrant locking (e.g. transport instrumentation calling Sentry.metrics.*)
+      return self if @mutex.owned?
+
       @mutex.synchronize do
         return unless ensure_thread
 

sentry-ruby/spec/support/shared_examples_for_telemetry_event_buffers.rb

@@ -122,6 +122,42 @@
     end
   end
 
+  describe "re-entrancy protection" do
+    let(:max_items) { 3 }
+
+    it "does not deadlock when add_item is called re-entrantly from send_items" do
+      reentrant_calls = 0
+
+      allow(client).to receive(:send_envelope) do
+        reentrant_calls += 1
+        # Simulate instrumentation calling back into the buffer mid-send
+        subject.add_item(event)
+        # also simulate a second re-entrant call to be sure
+        subject.add_item(event)
+      end
+
+      expect {
+        3.times { subject.add_item(event) }
+      }.not_to raise_error
+
+      expect(reentrant_calls).to be >= 1
+    end
+
+    it "silently drops the re-entrant item rather than raising" do
+      items_sent = []
+
+      allow(client).to receive(:send_envelope) do |envelope|
+        items_sent << :sent
+        subject.add_item(event)  # re-entrant; must be dropped, not raise
+      end
+
+      3.times { subject.add_item(event) }
+
+      expect(items_sent).not_to be_empty
+      expect(string_io.string).not_to include("deadlock")
+    end
+  end
+
   describe "error handling" do
     let(:max_items) { 3 }