Kimi-K2-Thinking native tool calling format

[KiruyaMomochi]

The implementation might support Kimi-K2-Instruct too, but I don't have enough disk space to test now :(

Almost silly copy-paste from DeepSeek V3.1 #15533, modified according to https://github.com/MoonshotAI/Kimi-K2/blob/main/docs/tool_call_guidance.md: matching function id instead of plain function name.

Considerations:

1. The official template does not contain any <think> tag at the end, so thinking_forced_open is false. Should we test it by modify the template manually?
2. Did not add template update instruction to models/templates/README.md for now, because their template has tojson(separators=(',', ':')). Although the value of separators is the same as default value, but we must remove it to make the template work for minja.
3. DeepSeek V3.1 might be possible to generate <｜tool▁calls▁begin｜>tool... and ignoring <｜tool▁call▁begin｜>, but I have not observed such behavior in Kimi-K2-Thinking and always get <|tool_calls_section_begin|><|tool_call_begin|>, therefore I'm removing the ? in the function regex.
   

   llama.cpp/common/chat.cpp

   Line 1751 in c4abcb2

   static const common_regex function_regex("(?:<｜tool▁call▁begin｜>)?([^\\n<]+)(?:<｜tool▁sep｜>)");

   
   Actually, I always get an extra <|tool_calls_section_end|> when keeping ?, but I have not been able to fix it, so finally removed the ?.
4. Have not tested lower quantized variants, maybe they could have different behavior which need to adapt the current parser?

For maintainers: I may have a busy weekend so fell free to edit directly if I'm not able to reply in time.

Closes #17155.

[calvin2021y]

#16932

[KiruyaMomochi]

Thanks! Will try #16932.

[KiruyaMomochi]

Finally get some to test and it worked well!

[KiruyaMomochi]

Close in favour of #16932.