convert_hf_to_gguf: add JoyAI-LLM-Flash tokenizer hash mapping to deepseek-v3

[dranger003]

• adding hash for jdopensource/JoyAI-LLM-Flash mapping to existing deepseek-v3
• DeepseekV3ForCausalLM architecture already supported
• moved GLM-4.7-Flash entry together with the other glm entries
• conversion and quantization successful and output generation looks great

[CISC]

Thanks, but convert_hf_to_gguf_update.py needs to be run to update convert_hf_to_gguf.py, which is why CI failed. :)

Anyway, I think I prefer that you create a new pre-tokenizer name for this in case there are any subtle differences discovered later on; move it out of pre_computed_hashes and add it in llama-vocab.cpp together with DEEPSEEK3_LLM and HUNYAN_DENSE.

[dranger003]

Ah, of course I changed the order after running the script, thanks.

[CISC]

You also need to add a section like this:

llama.cpp/src/llama-vocab.cpp

Lines 2039 to 2042 in 7af1cce

	} else if (
	tokenizer_pre == "hunyuan-dense") {
	pre_type = LLAMA_VOCAB_PRE_TYPE_HUNYUAN_DENSE;
	clean_spaces = false;