Add tsh command to resolve a single host

[rosstimothy]

tsh resolve allows identifying a single host either directly by hostname, or via custom search/predicate expression from a matched proxy template. The main use case is to provide a simple command for users that wish to use Match exec in their SSH config. Today, all matching in SSH config must be done via DNS and requires users to add some Teleport specific suffix/prefix, or use a wildcard entry to invoke a tsh proxy command.

For example, the SSH config generated today via tsh config is the following:

# Common flags for all local.dev hosts
Host *.cluster-name proxy.example.com
    UserKnownHostsFile "/Users/tim/.tsh/known_hosts"
    IdentityFile "/Users/tim/.tsh/keys/proxy.example.com/tim"
    CertificateFile "/Users/tim/.tsh/keys/proxy.example.com/tim-ssh/cluster-name-cert.pub"

# Flags for all local.dev hosts except the proxy
Host *.cluster-name
    Port 3022 !proxy.example.com
    ProxyCommand "tsh" proxy ssh --cluster=cluster-name --proxy=proxy.example.com:443 %r@%h:%p

This allows connections to the Teleport SSH service without using tsh directly via ssh foo.cluster-name. However, when migrating to Teleport, that requires the user to alter their existing workflow to include the cluster-name suffix. To remedy this, users can now augment their SSH config to utilize Match exec instead of globbing on the cluster name suffix with the following:

Match exec "tsh resolve -q %h"
    ProxyCommand "tsh" proxy ssh --cluster=cluster-name --proxy=proxy.example.com:443 %r@%h:%p

By default tsh resolve will output the matching host, if one was found, but if the -q flag provided like in the example above the output will be silenced. If no matches are found, or multiple matches are found, tsh resolve will exit with a non-zero exit code as per the Match exec requirements. If and only if a single host is resolved will tsh resolve exit with a zero exit code.

There are performance concerns that need to be taken into account before users adopt this in their SSH config. First, this may cause tsh resolve to be invoked on any SSH request. For example, even doing git pull/git push may now first require tsh resolve to exit with a non-zero exit code before interact with the git remote. Additionally, when tsh resolve finds a match, any connections to the node will require two connections to the cluster and two ListUnifiedResourcesRequests to resolve the host since the invocation of tsh resolve and tsh proxy ssh or tbot proxy ssh do not share any resources. For the reasons mentioned above, the SSH configuration generated by tsh config was not updated to include this new command. If users want to opt into this behavior they must acknowledge the latency concerns by manually editing the config.

changelog: Add a resolve command to tsh that may be used as the target for a Match exec condition in an SSH config.