[None][ci] split speculative test case into several small cases (NVIDIA#9209)

Signed-off-by: junq <[email protected]>

Author: QiJune

tests/integration/test_lists/test-db/l0_h100.yml

@@ -23,7 +23,8 @@ l0_h100:
   - unittest/_torch/modules
   - unittest/_torch/multimodal
   - unittest/_torch/sampler
-  - unittest/_torch/speculative
+  - unittest/_torch/speculative -k "eagle3"
+  - unittest/_torch/speculative -k "not eagle3"
   - unittest/_torch/thop/parallel
   - unittest/_torch/thop/serial
   # Only key models in H100: llama/mixtral/nemotron/deepseek