Enable jamba and kimi-linear for architecture tests

[zhaixuejun1993]

This pull request improves the handling of non-contiguous input tensors in split models for the OpenVINO backend and refines error and warning reporting in the decoder logic. The main changes include introducing a utility to reconstruct non-contiguous input tensors as contiguous OpenVINO tensors, updating operator support checks, and cleaning up debug output.

Non-contiguous input tensor handling:

• Added the function make_contiguous_split_input_tensor to reconstruct non-contiguous split model inputs into contiguous OpenVINO tensors, ensuring correct data layout for inference (ggml/src/ggml-openvino/utils.cpp).
• Updated convert_ggml_input_to_ov to use the new utility for any non-contiguous tensor in split models, replacing the previous custom logic for PERMUTE ops and extending support to other non-contiguous cases (ggml/src/ggml-openvino/utils.cpp).

Operator support and validation:

• Added has_non_contiguous_view_input to check for non-contiguous view inputs, and updated operator support logic to reject RMS_NORM ops with such inputs, preventing unsupported execution paths (ggml/src/ggml-openvino/ggml-openvino.cpp). [1] [2]

Error and warning reporting cleanup:

• Commented out or removed several std::cout and OPENVINO_ASSERT statements in compute_node_dynamic_dims to reduce console noise and avoid unnecessary assertion failures, making logs cleaner and less intrusive (ggml/src/ggml-openvino/ggml-decoder.cpp). [1] [2] [3] [4]## Overview

Additional information

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: