agent: Improve Pantheon support

[shazow]

Support for https://github.com/PegaSysEng/pantheon nodes in the vipnode agent.

Bulk of the work would be implementing support in the ethnode module: https://github.com/vipnode/vipnode/tree/master/ethnode

We'll need to make an implementation of the EthNode interface for the Pantheon RPC. This likely requires some non-web3-standard RPC calls, such as for whitelisting. Pantheon will need to support this.

• Confirm that Pantheon has the necessary RPC methods available to implement EthNode. If not, work with the Pantheon team to add them.
• Implement EthNode interface for Pantheon nodes.
• Confirm it's working. 👨‍🏭

Nice to haves:

• Add CheckCompatibility to make sure admin_* RPC is enabled with appropriate instructions.
• Automatically find Pantheon node key ($PANTHEON/key, includes 0x prefix)

[shazow]

Did a bit of looking at Pantheon docs:

• Add/Remove Peer are there ☑️
• Add/Remove trusted or reserved peer are not really there. They have whitelisted peers, but sounds like that restricts all other peers? What we really need is to add peers in a way that bypasses the internal limits (which is how geth's trusted peers and parity's reserved peers work) so that said inbound peers are able to connect even if the node is otherwise "full" of normal peers.
  Update: Chatted with the Pantheon team on Gitter, sounds like admin_addPeer does both outgoing and incoming connection whitelisting (and also retrying), so it should do the trick.

[shazow]

Turns out there's no way to run Pantheon with default RPCs exposed to the public and admin RPC accessible locally-only (such as using IPC/unix sockets on geth).

Related gitter discussion: https://gitter.im/PegaSysEng/pantheon?at=5d113b8330cc05479fbb623f

Related issue on their end: https://pegasys1.atlassian.net/browse/PAN-2839

This is fine if you don't need to expose any RPCs to the public, then you can --rpc-ws-api=ETH,NET,WEB3,ADMIN bound to 127.0.0.1 and the vipnode agent will use it from the same machine. But if you need to access the public node RPCs from an unauthenticated environment, then we can't have both.

Another alternative is to write a small JSONRPC2 muxer that just whitelists methods (I have some of that code in-progress, might repurpose it beyond mini-Infura) to workaround this limitation. Not a clean solution, but could be an acceptable temporary workaround.

For now, we'll have to assume that admin_* is enabled over websocket RPC. We should add an extra check/error with instructions if this is not the case.