Refactoring LLM Message Display for Trace Readability

[murugnn]

This enhancement focuses on improving the readability of LLM conversation history within trace and span inputs/outputs.

Previously, the system only extracted and displayed the last message from conversation arrays (like those used by OpenAI and LangChain). This made it difficult to review the full context of multi-turn interactions.

The update to the prettifyMessage utility now extracts all individual messages from recognized array formats. It structures them into a single string using:

A Message Header (---[ ROLE MESSAGE ]---) to identify the message type (e.g., HUMAN, AI, tool_call).

A Message Divider (\n\n ----------------- \n\n) to clearly separate each message.

This structured string is consumed by the front-end rendering components, which interpret the delimiters to render each message as a separate, collapsible section. This allows users to easily navigate and review the complete conversational flow within a single trace or span.

[dsblank]

This has been fixed with recent updates to the Pretty mode. Please reopen if you find any issues. Thanks for reporting!