Add tokio runtime precondition guard to all public sync entry points on EtcdCoordinator

[coderabbitai]

Summary

Only connect() and status() on EtcdCoordinator (in crates/gossip-coordination-etcd/src/backend/coordinator.rs) currently enforce the runtime precondition guard:

assert!(
    tokio::runtime::Handle::try_current().is_err(),
    "connect() must not be called from within an active Tokio runtime"
);

All other public sync entry points that reach self.runtime.block_on() are missing this guard. Callers invoking these methods from an async context will deadlock inside Tokio instead of receiving the explicit contract error documented on the type.

Affected methods

• etcd_get
• etcd_txn
• etcd_lease_grant
• etcd_lease_keep_alive_once
• etcd_lease_revoke
• list_active_runs_into
• gc_stale_initializing_runs_into
• Sync trait implementations backing create_run, renew, and related operations

Proposed fix

Add the same assertion at the top of each affected method, before any use of self.runtime or block_on, mirroring the existing connect()/status() semantics.

Background

Deferred from PR #148 (module split refactor) to keep that PR focused. See the review comment: #148 (comment)

Requested by: @ahrav