feat: feishu interactive card approval for agent permission requests

[xinuxZ]

When an agent requests human approval (e.g. to execute a shell command), send an interactive Feishu card with Approve/Reject buttons instead of requiring the user to type /approve manually.

Key changes:

• Approval notification: ApprovalManager broadcasts new requests via tokio broadcast channel; bridge layer subscribes and pushes cards to the last active Feishu user in real time

• Concurrent message dispatch: each inbound message now spawns its own task so a long-running agent call (blocked on approval) no longer prevents the user from sending /approve

• Feishu card callback parsing: support both card.action.trigger and application.bot.menu_v6 event types; button clicks are converted to /approve or /reject commands

• Feishu WebSocket receive mode: full implementation with reconnect backoff, protobuf frame decoding, ping/pong heartbeat, and multi-frame message reassembly

• exec_policy smart approval: allowlisted commands skip approval (fast path); unlisted commands escalate to approval instead of hard-blocking; approved commands are persisted to config.toml

• Hand agent model restore: use tag-based detection instead of hardcoded name; always re-apply default_model on DB restore

• Cross-compilation fix (openfang-cli/Cargo.toml, Cross.toml): add openssl vendored feature and rustls-tls for reqwest to fix cross build --target x86_64-unknown-linux-musl failure — the musl Docker container lacks system OpenSSL, so vendored mode downloads and statically compiles OpenSSL from source

Summary

Changes

Testing

• cargo clippy --workspace --all-targets -- -D warnings passes
• cargo test --workspace passes
• Live integration tested (if applicable)

Security

• No new unsafe code
• No secrets or API keys in diff
• User input validated at boundaries