tool-call: fix DeepSeek R1 Qwen distills

[ochafik]

Fixes tool call support of DeepSeek-R1-Distill-Qwen-7B & 32B (follow up to #9639), and adds <think>thoughts</think> parsing.

(Split off Minja changes in #11641, will declutter diff once merged)

• Had to work around the official template:
  • It doesn't describe the available tools, and the backfill done by Minja wasn't phrased well enough (for the 7B model), so I've added autogenerated tool call examples to minja's revamped "polyfill" behaviour (using a delta template eval).
    sync: minja #11641
  • It ignores message.tool_calls if message.content is not null, updated / testing the server output accordingly (better oai compliance)
  • After a tool result, it leaves the prompt hanging on a <｜tool▁outputs▁end｜> instead of ending with <｜end▁of▁sentence｜><｜Assistant｜>).
    • Hacked a workaround so the default template now works well with this branch
    • Added / documented better template (models/templates/llama-cpp-deepseek-r1.jinja)
• Both 8B & 32B models seem to take liberties with their tool call start tag, so accepting variations of the syntax (which then triggers the lazy grammar / full compliance)
• I've added a "thoughts" field for <think> content to the API similar to the "tool_plan" output of tool-call: support Command R7B (+ return tool_plan "thoughts" in API) #11585
  • Note: thoughts from previous messages are explicitly stripped from the prompt by the template, and we don't try and force them back (template wouldn't render tool calls if there was any content anyway, and while stripping thoughts will locally reset the KV cache, losing the hot tokens of the tool calls, it will save up tokens a lot in the long term over a chat)
• Added the Q4_K_M quant to some (but not all) slow server tests (had to tell it not to overthink, bit... ironic).
• Added slow tool result server tests (checking models make some use of tool call results, which some struggle a bit with)

# Please try with and w/o the chat template override and report back on your results :-)
llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M \
  --chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q6_K_L \
  --chat-template-file models/templates/llama-cpp-deepseek-r1.jinja

TODOs:

• Test content is null for tool calls
• Test changes in Minja repo (add backfill goldens, maybe start adding control over backfill behaviour?)
• Test prompting w/ tool results in server tests
• send --jinja --chat-template chatml support separately (tool-call: allow --chat-template chatml w/ --jinja, default to chatml upon parsing issue, avoid double bos #11616)
  • send double BOS fix separately
• send opportunistic fixes for Command R7B separately (tool-call: command r7b fix for normal responses #11608)

Possible follow ups

• Document differences between stream & non-stream modes (thought & tool_plan not sent in stream)
• look at the Llama distill more closely (see Eval bug: trivial grammar crashes (DeepSeek R1 Distill Llama 8B) #11591)