Explicitly test diagnostic output

To verify the changes in #4, I intentionally broke Diagnostic() by
prefixing lines with "!" instead of "#" but the tests still passed.
The spec[1] allows a test harness to silently ignore lines it doesn't
understand, so that behavior makes sense.

Red light green light testing[2] requires that our test be capable of
failing if something is broken.  This commit adds that by comparing
the actual output against our expected output.

Hat tip to @wking for suggesting to compare output[3].

1: http://testanything.org/tap-version-13-specification.html#anything-else
2: http://stackoverflow.com/a/404860/174463
3: mndrix/tap-go#4 (comment)

Author: mndrix

util/tap/test/diagnostic/main.go

@@ -1,13 +1,37 @@
 package main
 
-import "github.com/mndrix/tap-go"
+import (
+	"bytes"
+	"io"
+	"os"
+
+	tap "github.com/mndrix/tap-go"
+)
 
 func main() {
+	// collect output for comparison later
+	buf := new(bytes.Buffer)
 	t := tap.New()
+	t.Writer = io.MultiWriter(os.Stdout, buf)
+
 	t.Header(1)
 	t.Diagnostic("expecting all to be well")
 	t.Diagnosticf("here's some perfectly magical output: %d %s 0x%X.", 6, "abracadabra", 28)
 	t.Diagnostic("some\nmultiline\ntext\n")
 	t.Diagnosticf("%d lines\n%s multiline\ntext", 3, "more")
-	t.Pass("all good")
+
+	got := buf.String()
+	t.Ok(got == expected, "diagnostics gave expected output")
 }
+
+const expected = `TAP version 13
+1..1
+# expecting all to be well
+# here's some perfectly magical output: 6 abracadabra 0x1C.
+# some
+# multiline
+# text
+# 3 lines
+# more multiline
+# text
+`