feat(cli): add node peers, mempool, and metrics subcommands

[ByteYue]

Summary

Thin wrappers over reth's existing JSON-RPC so operators can ask the common health questions without remembering the RPC method names:

• node peers [--detail] → admin_peers. Default view shortens each peer into a single readable line (id prefix, inbound/outbound, client name, enode prefix); --detail dumps each entry as JSON; --output json emits the raw array.
• node mempool [--content] → txpool_status, optionally combined with txpool_content grouping pending tx counts by sender.
• node metrics → aggregates chain_id, block_number, peer_count, txpool_pending / queued, and eth_syncing into one snapshot. Each sub-query has an independent 3s timeout; failures become ? / null fields so the command still works on nodes that expose only a subset of RPC namespaces.

Added a -32601 / Method not found hint in errors.rs pointing at the node's --http.api list — admin_* / net_* typically require opt-in that users don't always know about.

No node-side changes

Every RPC method used is already exposed by reth when the relevant namespaces are enabled on the execution node.

Test plan

• cargo build -p gravity_cli --profile quick-release clean with RUSTFLAGS=\"--cfg tokio_unstable\"
• node metrics against live localnet → chain_id=31337, block_number=676, txpool_pending=0, syncing=no displayed correctly; peer_count=? gracefully when net_peerCount not exposed
• node metrics --output json → structured output with null for unexposed methods
• node mempool against live localnet → pending: 0 / queued: 0
• node peers against node without admin namespace → clean error + new hint
• node metrics against unreachable RPC → all ? fields, exit 0 (defensive)
• node mempool without --rpc-url → clean required-flag error with init hint

[github-actions]

This PR is stale because it has been open 45 days with no activity. Remove the stale label, comment or push a commit - otherwise this will be closed in 15 days.