Add ONNX and TensorRT export for LT-DETR instance segmentation

[yutong-xiang-97]

What has changed and why?

Add ONNX and TensorRT export support for LT-DETR instance segmentation.

Note: it now exports both images and orig_target_size inputs so predicted boxes and masks can be scaled to the caller-provided image size - something listed as TODO in object detection. The TensorRT helper now configures matching optimization profiles for additional inputs with dynamic batch dimensions, such as orig_target_size, while rejecting unsupported dynamic non-batch dimensions.

How has it been tested?

1. Unit tests covering

• static and dynamic batch export
• FP16
• simplification
• opset selection (ranging from 16 to 20 on CI)
• model aliases and non-square inputs
• prediction parity

3. A vibe-coded verification script that checks:

• Both inputs have a dynamic batch dimension.
• TensorRT accepts configured batch sizes, resolves output shapes, and produces finite outputs with the expected dimensions.
• TensorRT and ONNX Runtime produce compatible detections for identical random inputs.
  Detections are aligned order-independently using labels, boxes, and scores.
• Boxes and scores are numerically close after alignment.
• Labels match exactly.
• Raw mask-logit error statistics are reported.
• Thresholded masks are compared using pixel agreement, IoU, and agreement away from the zero-logit boundary.

Did you update CHANGELOG.md?

• Yes
• Not needed (internal change)

Did you update the documentation?

• Yes
• Not needed (internal change without effects for user)

[yutong-xiang-97]

/review

[liopeer]

LGTM! In the future we should not have orig_target_size in the exported model anyway. The model graph should only forward the image and yield class logits and box coordinates, working on this in #860, but this will take some time (and likely a design doc as well).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3fa4dc6f1c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9d144269d8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d0b96bcac2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".