Efficient SAM is not compliant with onnx standard

[DaniAffCH]

The model image_segmentation_efficientsam introduced in #258 does not comply with the ONNX standard.

To verify this, you can use the following snippet:

import onnx
model = onnx.load("image_segmentation_efficientsam_ti_2024may.onnx")
onnx.checker.check_model(model, full_check=True)

The ONNX checker raises the following exception:

onnx.onnx_cpp2py_export.shape_inference.InferenceError: [ShapeInferenceError] Inference error(s): (op_type:Tile, node name: /mask_decoder/Tile): [ShapeInferenceError] Inferred shape and existing shape differ in rank: (5) vs (0)
(op_type:Reshape, node name: /mask_decoder/transformer/Reshape): [ShapeInferenceError] Dimension could not be inferred: incompatible shapes