feat: return structured suggestion objects with display label and full prompt

[Ridwannurudeen]

Summary

• Suggestions now return {"display": "...", "prompt": "..."} objects instead of plain strings
• display is the short 2-5 word button label shown in the UI
• prompt is a full, detailed message sent to the agent when the button is clicked
• Includes graceful fallback: legacy plain strings are returned as display=prompt

Motivation

The suggestion buttons currently show a short label (e.g. "Analyze ETH trend") and that same short string is what gets sent to the agent. A richer, more detailed prompt improves the quality of agent responses significantly. For example, clicking "Analyze ETH trend" now sends: "Analyze Ethereum's price trend over the past 30 days, including volume, momentum, and key support and resistance levels."

Changes

api/api_types.py

• Add Suggestion dataclass with display: str and prompt: str

templates/suggestions.jinja2

• Updated to generate JSON objects with both display and prompt fields
• Added detailed example responses showing the expected format

server/fastapi_server.py

• Import Suggestion from api_types
• Update handle_suggestions_request return type from List[str] to List[Suggestion]
• Parse structured objects from LLM response; fall back gracefully for plain strings

Test plan

• Verify /api/v2/agent/suggestions returns {"suggestions": [{"display": "...", "prompt": "..."}, ...]}
• Verify frontend receives display for button label and prompt as the message to send on click
• Verify graceful fallback when LLM returns plain strings (display == prompt)

Closes #40

[Ridwannurudeen]

Hi team! Just a friendly ping on this PR. It returns structured Suggestion objects (with display_label and full_prompt fields) instead of raw strings, making it easier for callers to render suggestions in UI without re-parsing. No breaking changes — happy to adjust anything if you have feedback.