feat(api): report reasoning token usage

[sufubao]

Summary

• add completion_tokens_details.reasoning_tokens to Chat Completions usage
• count generated token IDs through the configured reasoning parser for streaming and non-streaming responses, including n-choice aggregation
• propagate the count to Responses API output_tokens_details.reasoning_tokens
• document the response fields in the English and Chinese reasoning parser guides

Motivation

LightLLM already separates reasoning text, but its usage model did not expose the number of reasoning tokens. Downstream gateways therefore reported reasoning_tokens as zero even when the response contained reasoning.

This follows the current vLLM main design: retain generated token IDs, let the active reasoning parser count them, and populate completion token details only when a reasoning parser is configured.

Verification

• black --check --line-length 120 on all changed Python files
• flake8 on all changed Python files