feat(server): encode DeepSeek V4 reasoning effort - #630

Summary

Add the official DeepSeek V4 Flash low, high, and max reasoning-effort encodings to the native renderer.

• low adds no model-facing prefix
• high and max use the official prompt prefixes
• accepts reasoning.effort, top-level reasoning_effort, and chat_template_kwargs.reasoning_effort
• accepts the official chat_template_kwargs.thinking switch
• maps medium to high and x-high/xhigh to max for client compatibility
• advertises the DeepSeek-specific effort set through /props
• preserves the normalized effort when FlowKV rerenders a request

Official encoding reference: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/main/encoding/README.md

Published qualification patches:

• lucebox-deepseek-v4-reasoning.patch
• lucebox-deepseek-v4-official-effort-v2.patch

Validation

• Added renderer and /props unit coverage
• Compiled the current-main renderer and its Jinja dependencies locally, then ran focused assertions for low/high/max and thinking-disabled behavior
• git diff --check