UR-Nexus 1.87.5

• Fixed GPT-6.1 Terra requests failing with HTTP 400 because UR sent the
  deprecated max_tokens field. Direct OpenAI Chat Completions now use
  max_completion_tokens independently of cached reasoning metadata, including
  newly discovered models. Compatible servers retain their own token contract.
• Added GPT-6.1 Terra's documented reasoning levels, default, context/output
  limits and Fast/Flex controls. Agent requests automatically use Responses,
  as required for this model's tool calling, with max_output_tokens.
• Added credential-free regressions for Terra tool calls and continuation,
  full-query streaming, thinking off/on and serving tiers. Updated provider,
  validation and installation documentation for this release.