improve diff for []byte values

[rogpeppe]

[]byte is commonly used to encode strings, so it's hard to read the diff when it's phrased in terms of hex bytes rather than strings.

For example, compare the two diff outputs in this example:

  []string{
- 	"hello",
  	"foo",
+ 	"foo",
  	"barbaz",
- 	"blahdieblah",
+ 	"added",
+ 	"here",
+ 	"hrmph",
  }

  [][]uint8{
- 	{0x68, 0x65, 0x6c, 0x6c, 0x6f},
  	{0x66, 0x6f, 0x6f},
+ 	{0x66, 0x6f, 0x6f},
  	{0x62, 0x61, 0x72, 0x62, ...},
+ 	{0x61, 0x64, 0x64, 0x65, 0x64},
+ 	{0x68, 0x65, 0x72, 0x65},
- 	{0x62, 0x6c, 0x61, 0x68, 0x64, 0x69, 0x65, 0x62, 0x6c, 0x61, 0x68},
+ 	{0x68, 0x72, 0x6d, 0x70, 0x68},
  }

I think it would be better to see something like this instead:

  [][]byte{
- 	[]byte("hello"),
  	[]byte("foo"),
+ 	[]byte("foo"),
  	[]byte("barbaz"),
- 	[]byte("blahdieblah"),
+ 	[]byte("added"),
+ 	[]byte("here"),
+ 	[]byte("hrmph"),
  }

[dsnet]

The output doesn't have to be valid Go (it already isn't), we could probably simplify it:

  [][]byte{
- 	"hello",
  	"foo",
+ 	"foo",
  	"barbaz",
- 	"blahdieblah",
+ 	"added",
+ 	"here",
+ 	"hrmph",
  }

[rogpeppe]

It would be quite nice if that were valid Go.

[rogpeppe]

FWIW I just had an idea that the output might be nicer if we used a Transform step to translate []byte to string
(https://play.golang.org/p/nlUN8xivsef), and it is but not by much. I wonder if it might be better if all the diff output looked the same:

  [][]uint8{
- 	{0x68, 0x65, 0x6c, 0x6c, 0x6f},
  	Inverse(bytesTostring, string("foo")),
+ 	{0x66, 0x6f, 0x6f},
  	Inverse(bytesTostring, string("barbaz")),
- 	Inverse(bytesTostring, string("blahdieblah")),
+ 	Inverse(bytesTostring, string("added")),
+ 	{0x68, 0x65, 0x72, 0x65},
+ 	{0x68, 0x72, 0x6d, 0x70, 0x68},
  }