[FEATURE] Text-to-GEST parser for structured scene graph extraction

[ncudlenco]

Is your feature request related to a problem? Please describe.
Currently there is no automated way to convert free-form text into a structured GEST (scene graph) representation in mta-sim. This limits the ability to leverage textual data for scenario generation, annotation, or downstream processing that relies on formal scene graphs. Manual conversion is time-consuming and can introduce inconsistencies.

Describe the solution you'd like
Implement a text-to-GEST parser that:

• Accepts free-form or semi-structured text as input.
• Uses formal GEST rules (to be defined based on the GEST paper) to parse and extract entities, relationships, actions, and temporal information.
• Outputs a structured JSON representation of the scene graph (GEST).
• Clusters the text by regions/locations, integrating these segments into the final graph while preserving temporal relationships.
• Is modular and extensible to accommodate updates to GEST rules or new entity types.

Describe alternatives you've considered

• Manual annotation and conversion (not scalable)
• Heuristic-based scripts with no formal rule integration (less robust and error-prone)

Acceptance Criteria

• Text-to-GEST parser is implemented
• Parser uses formal GEST rules for extraction
• Output is a structured JSON representation of the GEST scene graph
• Text is clustered by regions/locations and properly integrated
• Temporal relationships are preserved in the output
• Documentation describes parser usage and GEST rule format
• Tests verify parser correctness and integration
• No existing functionality is broken
• Performance impact is acceptable

Additional context

• GEST rule specification will be extracted separately from the GEST paper.
• Consider integration with the story teller system to generate text, GEST pairs