Incorrect non-unicode file path processing

[imeoer]

Description

If the image layer includes non-unicode file path, stargzify can't process correctly, example:

mkdir ./test
touch ./test/$(printf '\xFF')
touch ./test/$(printf '\xAA')
tar -czvf ./test.tar.gz ./test
stargzify file:test.tar.gz file:test.stargz
tar -xzvf test.stargz stargz.index.json

The index file stargz.index.json has two entries with the same name "�":

{
	"version": 1,
	"entries": [
		{
			"name": "./test/",
			"type": "dir",
			"modtime": "2020-09-08T10:33:57Z",
			"mode": 509,
			"uid": 1000,
			"gid": 1000,
			"NumLink": 0
		},
		{
			"name": "./test/\ufffd",
			"type": "reg",
			"modtime": "2020-09-08T10:33:47Z",
			"mode": 436,
			"uid": 1000,
			"gid": 1000,
			"NumLink": 0,
			"digest": "sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
		},
		{
			"name": "./test/\ufffd",
			"type": "reg",
			"modtime": "2020-09-08T10:33:57Z",
			"mode": 436,
			"uid": 1000,
			"gid": 1000,
			"NumLink": 0,
			"digest": "sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
		}
	]
}

Possible Solution

Treat file path as []byte like xattr value, so that we can serialize the path as base64 encoded string to JSON index file, but we need to distinguish between plain path and base64 encoded path using extra entry field like "name_encoded": true.