MCP access part 11: tsh auto reconnect

[greedy52]

part of:

• Teleport MCP access MVP #54705

This PR attempts to improve user experience when Teleport MCP server is "disconnected" (network issue, expired tsh session).

Now tsh mcp connect by default starts a new remote MCP session if the previous one is interrupted by network issues or tsh session expiration. The behavior can be overwrite by --[no]-auto-reconnect flag.

TODO(s)

• make derive
• make -C integrations/operator crd
• make -C integrations/terraform docs

examples (click to expand)

`tsh mcp config`

$ tsh mcp config everything
Found MCP servers:
everything

Here is a sample JSON configuration for launching Teleport MCP servers:
{
  "mcpServers": {
    "teleport-mcp-everything": {
      "command": "/Users/stevehuang/go/github.com/gravitational/teleport/build/tsh",
      "args": ["mcp", "connect", "everything"],
      "env": {
        "TELEPORT_HOME": "/Users/stevehuang/.tsh-local"
      }
    }
  }
}

By default, tsh automatically starts a new remote MCP session if the previous
one is interrupted by network issues or tsh session expiration.
Auto-reconnection is recommended when MCP sessions are stateless across
requests. To disable it, use the --no-auto-reconnect flag. If disabled, you may
need to manually restart your client when encountering “disconnected” errors.

Tip: use --client-config=claude to update your Claude Desktop configuration.
You can also specify a custom config path with --client-config=<path> to update
a config file compatible with the "mcpServer" mapping.

Claude network issue

Claude expired tsh session

[gabrielcorado]

LGTM, just a few minor comments. Tested with Claude Desktop.

[backport-bot-workflows]

@greedy52 See the table below for backport results.

Branch	Result
branch/v18	Failed