[app] show only dataset attributes with values

### StatGPT Backend version

0.1.0

### What is the problem this feature will solve?

The app can't reliably answer vague data queries without indicator specification (e.g., "give me data from dataset X"), asking for clarifications.

### What is the feature you are proposing to solve the problem?

Sometimes datasets have attributes specifying the most important indicators they contain. These indicators are stored in attribute values. Therefore, a solution for such datasets could be:

- show dataset level attributes with their values (and only attributes with set values) in stages
- stage outputs are passed to agent context, and thus will be accessible to it to retrieve the most important indicators in response to vague queries
- NOTE: additional instructions regarding these attributes (e.g., attributes names, how to handle them properly) should be added to agent's system prompt

### What alternatives have you considered?

_No response_

Start by tracing how dataset-level attributes are collected, rendered in stages, and passed into agent context, then locate the agent system prompt handling. Confirm how set versus unset attribute values are represented. Done means staged outputs expose only valued attributes and the prompt provides instructions for using them in vague-query responses.

Written by the indexing model from the issue text.