Fix stream fallbacks for conversation responses

[nmaguiar]

Motivation

• Prevent runtime failures when usestream=true but the LLM provider implements promptStreamWithStats without promptStreamJSONWithStats, which broke JSON-structured conversation flows.
• Ensure showthinking=true uses safe non-stream code paths so thinking logs and legacy prompt methods remain compatible with providers lacking stream-JSON support.

Description

• Add a canStreamJson guard and use it to call promptStreamJSONWithStats only when that method exists, falling back to non-stream methods otherwise in the agent loop (mini-a.js).
• Harden the showthinking path to prefer existing non-stream JSON/raw methods and to avoid entering streaming code paths that are incompatible with thinking logs.
• Apply the same canStreamJson checks and fallbacks to the chatbot flow so chatbot-mode streaming behavior matches the agent flow.
• All changes are confined to mini-a.js and preserve streaming behavior when full stream APIs are available.

Testing

• Ran node -c mini-a.js to validate syntax and loading, which succeeded.
• Attempted ojob tests/miniAUtils.yaml to exercise the test suite, but it could not be executed in this environment because the ojob command is not available (test run failed with command not found).