Support yololite with fused NMS

[leeclemnet]

What does this PR do?

roboflow/yololite#7 and https://github.com/roboflow/roboflow-train/pull/763 extend the yololite trainer to export the same 3 ONNX variants as other yolos: (1) static batch size non-fused NMS, (2) dynamic batch size non-fused NMS, (3) 'dynamic' batch size fused NMS.

yololite already exports and registers (2), which inference_models has been using so far, this PR adds support for (1) and (3). The change is backwards compatible so already-trained and exported yololite still work (see testing).

This PR can be merged independently of https://github.com/roboflow/roboflow-train/pull/763.

Type of Change

• New feature (non-breaking change that adds functionality)

Testing

• I have tested this change locally
• I have added/updated tests for this change

Test details:

• Coin counting integration tests for the 3 different onnx variants that yololite will export after https://github.com/roboflow/roboflow-train/pull/763
• Tested previously exported yololite onnx model against local inference server in workflows - backwards compatible

• Tested new model version against local inference server in workflows - similar output

• Tested new model version against current serverless inference in workflows - still works, same output as local

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code where necessary, particularly in hard-to-understand areas
• My changes generate no new warnings or errors
• I have updated the documentation accordingly (if applicable)

Additional Context