Update sparse mesh production profile and FP32 contract

Normalize SubM SparseConv3D outputs to stable input coordinate order

Use direct fixed-cardinality TensorRT output bindings for SubM SparseConv3D

Promote Companion Forge ONNX TensorRT sparse custom-op ABI

Fix SparseConv3D descriptor to fixed-cardinality SubM dynamic N

Add production sparse-to-FlexiCubes TensorRT mesh topology plugin

Make SparseConv3D runtime robust by deriving SubM structure from native spconv weights

Add FlexiCubes mesh topology TensorRT numerical validation

Add isolated SparseConv3D TensorRT diagnostic

Fix SparseConv3D ONNX weight layout to native spconv O-K-K-K-I

Fix SparseConv3D ONNX weight layout to native spconv O-K-K-K-I

Handle TensorRT DDS output buffers through upper-bound aliases

Add numerical execution tests for custom TensorRT sparse ops

Fix TensorRT optimization profile API return handling

Fix TensorRT network input profile iteration

Add TensorRT parser/build validation for all custom ONNX ops

Use eager annotations required by TensorRT plugin registry

Use TensorRT-compatible companionforge plugin namespace

Use TensorRT-compatible plugin namespace while keeping ONNX custom domain

Fix ONNX TensorProto dtype for mesh topology graph

Add standalone ONNX custom-op graph generator

Add ONNX custom-domain builders for Companion Forge sparse ops

Add TensorRT 11.2 custom plugins for all AniGen sparse and topology ops

Add exact AniGen sparse custom-op reference implementations

Document remaining sparse custom operators

Add L4 ONNX TensorRT hybrid runtime scaffold