Every Kapacitor client (`github.com/influxdata/kapacitor/client/v1`)
instantiates its own `http.Transport`, unless one is provided. A
`http.Transport` will cache socket connections for future use, which
leaves connections open.
Chronograf creates a new Kapacitor client upon every request for several
endpoints. Thus each time these endpoints were hit, new connections
would be opened but never closed until Chronograf is shut down.
These changes pass a singleton `http.Transport` when creating new
Kapacitor clients, so that connections are reused while Chronograf is
running.