chore: log failure to run CLI command (#12863)

Log failure to run CLI command

Prior code only conditionally logged the output when `LOTUS_DEV` env
var was present. This doesn't make sense for a number of reasons:

* In case of an error starting up `lotus daemon` via systemctl for
  example the logs would miss the error. One then has to search through
  system journal to see what happened.

* There are no other places that I can find where `LOTUS_DEV` env var is
  used. The commit that introduced this condition is very old with no
  clear commit message to shed light into the rationale.

For the first reason alone, the changes here remove that condition, and
log the message when the logging output is not standard err or out. This
way we at least cover the case where lotus is run in a production
environment and would avoid partial logging of errors.

Author: masih

cli/helper.go

@@ -8,6 +8,7 @@ import (
 	"os/signal"
 	"syscall"
 
+	logging "github.com/ipfs/go-log/v2"
 	ufcli "github.com/urfave/cli/v2"
 )
 
@@ -46,11 +47,13 @@ func RunApp(app *ufcli.App) {
 	}()
 
 	if err := app.Run(os.Args); err != nil {
-		if os.Getenv("LOTUS_DEV") != "" {
-			log.Warnf("%+v", err)
-		} else {
-			fmt.Fprintf(os.Stderr, "ERROR: %s\n\n", err) // nolint:errcheck
+		if cfg := logging.GetConfig(); !(cfg.Stdout || cfg.Stderr) {
+			// To avoid printing the error twice while making sure that log file contains the
+			// error, check the config and only print it if the output isn't stderr or
+			// stdout.
+			log.Errorw("Failed to start application", "err", err)
 		}
+		_, _ = fmt.Fprintf(os.Stderr, "ERROR: %s\n\n", err)
 		var phe *PrintHelpErr
 		if errors.As(err, &phe) {
 			_ = ufcli.ShowCommandHelp(phe.Ctx, phe.Ctx.Command.Name)