Add read-only Solana token risk plugin

[guoqiangliu-ocean]

What changed

• add token-risk-check, a single-tool Solana WIT component
• query mint metadata and largest accounts through operator-configured Solana RPC
• query observed Solana pools through the current DEX Screener token-pairs API
• report authorities, Token-2022 extensions, concentration, and liquidity as compact red/amber/green evidence
• keep all assessment logic in a pure host-testable core with a thin wasi:http shim
• add offline mock tests, including malformed data, RPC failure, and prompt-injection fail-closed cases
• document T0 custody, threat model, configuration, limitations, and a worked example

Why

Autonomous agents need a read-only safety check before payment or swap tooling. This plugin provides that check without a private key, signing path, transaction builder, or raw RPC payload in the model context.

Safety

This is custody tier T0. The only tool argument is a base58 mint that must decode to 32 bytes. Endpoints come from the plugin's jailed operator config and must use HTTPS. Missing or malformed source data cannot produce a green result.

Validation

• cargo fmt --manifest-path plugins/token-risk-check/Cargo.toml --all -- --check
• cargo metadata --manifest-path plugins/token-risk-check/Cargo.toml --locked --no-deps --format-version 1
• python tools/build-registry.py --source-plugins plugins --check-metadata registry.json
• live shape check against Solana getAccountInfo and DEX Screener's documented /token-pairs/v1/solana/{tokenAddress} endpoint
• full repository workflow on the fork: host tests, clippy, wasm32-wasip2 release builds, registry packaging dry run, and the required validation gate all passed: https://github.com/guoqiangliu-ocean/zeroclaw-plugins/actions/runs/29940249752

The upstream run is awaiting the repository's standard first-time-contributor workflow approval; the identical workflow completed successfully on the fork.