Skip int8 model on GPU and merge from upstream

[c8s-wk]

• Added is_model_int8 to detect int8 models by filename or layer property.
• Refactored benchmarking to use run_benchmark_if_supported, which loads model files, checks for int8, and skips GPU benchmarking for int8 models with a clear message.
• Applies to both default and user-specified models(param).

[tencent-adm]

All committers have signed the CLA.

[github-actions]

The binary size change of libncnn.so (bytes)

architecture	base size	pr size	difference
x86_64	15657272	15657272	0 😘
armhf	6653616	6653616	0 😘
aarch64	9989032	9989032	0 😘

[Copilot]

Pull Request Overview

A concise summary of the purpose of the PR, followed by summarized bullets of changes

• Introduces logic to detect int8 models and skip GPU benchmarking when an int8 model is used
• Refactors benchmarking entry point to use run_benchmark_if_supported for both default and user-specified models
• Applies consistent skip behavior and messaging across model sources

Comments suppressed due to low confidence (2)

benchmark/benchncnn.cpp:52

• [nitpick] There are no tests verifying that GPU benchmarking is correctly skipped for int8 models; consider adding unit or integration tests for this branch.

    if (opt.use_vulkan_compute && strstr(comment, "int8") != NULL)

benchmark/benchncnn.cpp:54

• [nitpick] The parameter fixed_path is ambiguous; renaming it to something like is_user_specified_model or use_custom_model would clarify its purpose.

        if (!fixed_path)

[nihui]

Thanks for your contribution !