Possible ggml-1.0.0 compilation fixes, AI generated

[adamjrichter]

The attached small AI generated diff allowed me to compile stable-diffusion as of yesterday against newly released libggml-1.0.0, which
I understand removed some interfaces.

I do not currently understand the relevant code in either the stable diffusion call sites or the libggml callee sites to fully vouch for the patch. So, rather than making a pull request, I invite anyone who knows better to take over this patch and submit it or post a response to explain why the patch is bad. I hereby relinquish any copyright interest I may have in the two patches discussed in this message to the public domain. Perhaps later I may study the relevant bits of enough to be willing to submit it myself if nobody does this.

The AI said it also tried compiling against the version of ggml integrated with the stable diffusion source tree, and that that continued to compile, but I haven't checked that. I have tried generated a couple of images on a few models using sd-cli built with this patch, but I don't know if the relevant code actually executed. I may at least try to catch in gdb, but want to post now, because I am unlikely to have time to do so for the next half day.

As with previous versions of libggml, it is necessary to build it with C preprocessor symbol GGML_MAX_NAME defined to 160 or more. I've attached a second diff, this one against libggml, so that the include file matches how the library was actually compiled.

The ggml diff was produced by app.devin.ai with just instructions from me, and guidance on both diffs by gemini.google.com.

sd-ggml-1.0.0-compilation-fixes.diff.txt
ggml_max_name_160.diff.txt

[wbruna]

Related: #2354