CANN: add optional support for ACL Graph execution

[noemotiovon]

This commit adds support for executing ggml computational graphs using Huawei's ACL graph mode via the USE_ACL_GRAPH flag. The support can be enabled at compile time using the CMake option:

-DUSE_ACL_GRAPH=ON

By default, ACL graph execution is disabled, and the fallback path uses node-by-node execution.

Key additions:

• CMake option to toggle graph mode
• Graph capture and execution logic using
• Tensor property matching to determine whether graph update is required
• Safe fallback and logging if the environment variable LLAMA_SET_ROWS is unset or invalid

This prepares the backend for performance improvements in repetitive graph execution scenarios on Ascend devices.

[noemotiovon]

Model: Qwen2.5-0.5B

	with GGML_OP_CPY	with GGM_OP_SET_ROWS
with ACL Graph	fall back without ACL Graph	120tok/s
without ACL Graph	70tok/s	70tok/s

With ACL Graph Test：

user
Building a website can be done in 10 steps:
assistant
Certainly! Here are 10 steps to help you build a website:

1. **Define Your Purpose and Goals**
   - Identify what you want your website to do.
   - Determine your audience and what they want.

2. **Choose the Right Platform**
   - Select a web development platform that suits your needs.
   - Consider factors like stability, security, and scalability.

3. **Set Up a Domain Name**
   - Choose a domain name that is easy to remember and reflects your website.
   - Ensure it is available and easy to register.

4. **Choose a Domain Hosting Provider**
   - Select a reliable domain hosting service that meets your needs.
   - Consider features like speed, security, and scalability.

5. **Choose a Content Management System (CMS)**
   - Select a CMS that fits your needs and the complexity of your website.
   - Choose based on features, ease of use, and community support.

6. **Design Your Website**
   - Choose a design that is visually appealing and easy to navigate.
   - Consider the style, color scheme, and layout.

7. **Choose a Content Writer**
   - Hire a skilled content writer to create the content.
   - Consider the writer's skills, experience, and style.

8. **Choose a Website Builder**
   - Select a website builder that suits your needs.
   - Choose based on features, ease of use, and scalability.

9. **Implement Your Website**
   - Use your chosen content management system and website builder to create your website.
   - Test the website thoroughly for errors and usability.

10. **Launch and Promote Your Website**
    - Launch your website and start promoting it.
    - Engage with your audience through social media, email marketing, and other channels.

11. **Monitor and Update**
    - Monitor the performance of your website.
    - Keep your website updated with new content and features.

12. **Refine and Improve**
    - Regularly review and refine your website content and design.
    - Stay updated with the latest trends and best practices in web development.

These steps should help you build a website that meets your needs and helps drive traffic to your site.

> 
llama_perf_sampler_print:    sampling time =     117.75 ms /   469 runs   (    0.25 ms per token,  3982.91 tokens per second)
llama_perf_context_print:        load time =    7527.43 ms
llama_perf_context_print: prompt eval time =      40.20 ms /    20 tokens (    2.01 ms per token,   497.49 tokens per second)
llama_perf_context_print:        eval time =    3754.24 ms /   448 runs   (    8.38 ms per token,   119.33 tokens per second)
llama_perf_context_print:       total time =    4803.95 ms /   468 tokens
llama_perf_context_print:    graphs reused =        433

[hipudding]

Thank you for your contribution. Enabling acl_graph support greatly reduces NPU idle cycles, with particularly noticeable performance gains on models with smaller parameter sizes.
The code is basically fine; only a few minor changes are needed.

[hipudding]

I think use_cann_graphis not necessary. You need only set cann_graph_update_required = true for both acl_graph on and off.

[noemotiovon]

I think use_cann_graph is still needed. Even when cann_graph is enabled, we may fall back to eager mode if LLAMA_SET_ROWS is not set. So we need a way to track that cann_graph is not being used in this case.