Merge deepseek_v4_compressed_attention into dsv4_integration

Author: parambole

pytest.ini

@@ -24,6 +24,7 @@ addopts =
     --ignore=tests/unit/qwen3_next_vs_reference_test.py
     --ignore=tests/unit/deepseek32_vs_reference_test.py
     --ignore=tests/unit/engram_vs_reference_test.py
+    --ignore=tests/unit/deepseek_v4_vs_reference_test.py
 markers =
     tpu_only: marks tests to be run on TPUs only
     tpu_backend: marks tests that require a TPU-enabled JAX install (TPU PJRT plugin), but not TPU hardware