Proposal: Evidence Bench evidence-gated computational-science A2A sample

## Proposal

I would like to contribute Evidence Bench as a computational-science A2A ecosystem sample:

- source: https://github.com/kstawiski/evidence-gated-scientific-agent
- first public release: https://github.com/kstawiski/evidence-gated-scientific-agent/releases/tag/v0.3.0
- license: Apache-2.0
- SDK/protocol: official `a2a-sdk[fastapi]` 1.1.0, Agent Card advertising A2A 1.0 JSON-RPC

Evidence Bench demonstrates an evidence-gated scientific agent rather than a generic chat agent. Qwen performs planning, retrieval, and sandboxed Python/R analysis; Gemma independently audits methods and results; deterministic code validates tool policy, artifacts, claims, and provenance. The service supports isolated persistent workspaces, raw A2A file parts, bearer-authenticated task execution, MCP retrieval tools, immutable per-workspace PyPI/CRAN/Bioconductor environments, and report/provenance artifacts.

## Interoperability evidence

- repository A2A API tests pass in public CI;
- a deployed authenticated `SendMessage` task passed all 9 retrieval-grounding checks, including official SciPy/R sources, observed-URL enforcement, deterministic validation, and independent review;
- the Agent Card and JSON-RPC semantics are documented at https://github.com/kstawiski/evidence-gated-scientific-agent/blob/v0.3.0/docs/WEB_AND_A2A.md;
- the proposed contribution dossier is at https://github.com/kstawiski/evidence-gated-scientific-agent/blob/v0.3.0/docs/A2A_ECOSYSTEM_SUBMISSION.md.

Before preparing a significant PR, could maintainers advise which shape is preferred?

1. a small in-tree integration/sample that delegates to a running Evidence Bench instance; or
2. an external ecosystem/example entry linking to the full repository.

I am happy to reduce the contribution to the smallest useful A2A-specific example and follow the repository's test and documentation conventions.