fix(routing): force tool_choice=required in OneStepForwardReasoner to eliminate format drift

[ywang1110]

Due to format drift, LLMs may complete with finishReason=stop and return free text or JSON in Content instead of a structured function call, causing instruction parsing failures in OneStepForwardReasoner.

• Set tool_choice=required as a scoped state around the single completion call in OneStepForwardReasoner.GetNextInstruction, ensuring the LLM always returns the instruction as a function call and the state is cleaned up immediately after
• Extract GetChatCompletionsWithScopedState helper with try/finally to guarantee state cleanup even if the completion throws
• Parse response.FunctionArgs directly instead of Content, since tool_choice=required guarantees a function call response
• Apply ToolChoice in PrepareOptions after tools are rendered, guarded by !options.Tools.IsNullOrEmpty() to avoid API rejection on toolless agents

[qodo-code-review]

ⓘ You've reached your Qodo monthly free-tier limit. Reviews pause until next month — upgrade your plan to continue now, or link your paid account if you already have one.

[ywang1110]

@Oceania2018 Ready for review. Thank you!