Skip the decoding server entirely on pre-0.16 realtime

The decoding server was the QEC build's last consumer of the 0.16-era
realtime dispatch API on this branch: GpuRoceTransceiver.cpp stamps
cudaq_function_entry_t::routing_key, which the old cuda-quantum pinned
here does not have. Its CMake gates (realtime headers found; HSB + DOCA
libs found for the gpuroce component) never consulted
CUDAQ_REALTIME_FOR_0_16, so CI built it and failed.

Return early from decoding-server-cqr when the 0.16-era API is absent,
before the GPU RoCE detection, and pin CUDAQ_GPU_ROCE_AVAILABLE to
FALSE in the cache. Every dependent already guards on if(TARGET ...):
the core-server unit tests, the CQR plugin consumers (pymatching
realtime test, app examples, the link-closure guardrail), the
decoding_server tool, and its two-process test disable themselves with
the component. No dangling references remain.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: Ben Howe <bhowe@nvidia.com>

Author: bmhowe23

libs/qec/lib/realtime/decoding-server-cqr/CMakeLists.txt

@@ -13,6 +13,21 @@ else()
   set(_core_include_dir "${CMAKE_SOURCE_DIR}/libs/core/include")
 endif()
 
+# The decoding server is built on the 0.16-era cuda-quantum realtime dispatch
+# API (routing_key function-table routing, the device-graph scheduler).
+# Against an older cuda-quantum, skip the whole component: every dependent
+# target -- the gpuroce transport, the CQR plugin, the decoding_server tool,
+# and their tests -- is gated on if(TARGET ...) and disables itself with it.
+# TEMPORARY (see CUDAQ_REALTIME_FOR_0_16 in ../CMakeLists.txt).
+if(NOT CUDAQ_REALTIME_FOR_0_16)
+  set(CUDAQ_GPU_ROCE_AVAILABLE FALSE CACHE INTERNAL
+    "GPU RoCE transport component cudaq-qec-decoding-server-gpuroce available")
+  message(STATUS
+    "cudaq-qec-decoding-server: SKIPPED (requires the 0.16-era cuda-quantum "
+    "realtime dispatch API; see CUDAQ_REALTIME_FOR_0_16)")
+  return()
+endif()
+
 # ---------------------------------------------------------------------------
 # Optional: GPU RoCE transport via Hololink Sensor Bridge + DOCA
 #