[Enterprise] Add qualified OpenFold3 inference support

[yifeif-nv]

Parent epic: #998

Summary

Add a qualified OpenFold3 inference path through TensorRT Model Connect.

The implementation must cover the complete pinned inference configuration while documenting any feature generation, MSA, template, or other preprocessing that remains external to the native runtime.

Tasks

• Pin the exact OpenFold3 source repository, source revision, gated checkpoint revision, configuration, and applicable licenses or access requirements.
• Define the supported sequence, MSA, template, ligand, feature-generation, and preprocessing workflow.
• Define the output structure, confidence, ranking, and metadata contracts.
• Implement the required OpenFold3 model components and model-owned data transformations.
• Support the required Pairformer or trunk, recycling, diffusion, masking, and variable-length behavior.
• Package the inference components and required metadata into a reproducible Model Connect bundle.
• Add a native runtime example that does not require Python or PyTorch after bundle creation.
• Validate structural and confidence outputs against the pinned reference implementation using documented quality metrics and tolerances.
• Establish an initial BF16 qualification profile on at least one documented supported NVIDIA GPU.
• Benchmark representative input sizes for warmed steady-state latency, throughput where applicable, and peak memory against aligned eager PyTorch and torch.compile baselines, excluding engine-build time.
• Characterize maximum tested input sizes, valid-output rate, unsupported preprocessing, and known limitations.
• Add model-owned unit, integration, qualification, and end-to-end tests.
• Document clean-environment setup, gated-asset requirements, bundle generation, native execution, and result interpretation.

Acceptance criteria

A pinned OpenFold3 configuration runs reproducibly through a native TensorRT Model Connect bundle and produces structurally valid outputs within the agreed quality tolerances on representative inputs.

The qualified BF16 profile records the exact GPU and software stack and includes reproducible quality, validity, latency, and memory evidence, with all external preprocessing and gated-asset requirements documented.