Experimental IO JSON for logs & workflow communication protocols

[doxav]

This draft pull request follows discussion on logs integration and multi-agent workflow communication in JSON.

It introduces a suite of utilities for importing, exporting, and converting Trace graphs in JSON format (a Trace‑Graph JSON (TGJ) format, and OpenTelemetry (OTel)).

This includes: TGJ ingestion, TGJ export, and Otel-to-TGJ conversion.

This will allow interoperability with logs/telemetry systems and simplify the process of linking JSON-based traces to executable code (aim MCP, A2A, ACP protocols).

• Added a detailed README.md for the opto/trace/io directory, describing the purpose of the new utilities, supported modules (tgj_ingest.py, tgj_export.py, otel_adapter.py), usage examples, and test coverage.

TODO

• Add a meaningful demo of MCP, ACP, and A2A protocols usage ?
  The challenge of the demo is to define what kind of optimization or modification of trainable parameters should be illustrated, and how.
  The JSON graph IO will provide a new mean of observability, but the optimization update mechanism may only operate locally at this stage (within the same process as trace runs).
  Unless another mechanism is defined for updating external trainable parameters, I think this is out of scope ?
• Maybe we could focus the first demo on MLFlow telemetry if it can enrich an existing MLFlow trace optimization process ?