Expand the recording of the original string for some literals

Previously we recorded the author's original string for floats, as gren-format had no idea of how the original author formatted them.
It turns out we probably need to do that for more types of literals.

I discovered this for string literals that contain unicode escapes; the parser coverts to those unicode escapes to actual utf-8 sequences, and gren-format doesn't know the original representation of them. This happens for hex integers too, although, this somehow feels less important of a concern.

Given this input:

module Repro exposing (..)


emoji =
    "\u{1F600}"


letterA =
    '\u{0041}'


carriageReturn =
    "\u{000D}"


hexLower =
    0xff


hexPadded =
    0x00FF


floatKept =
    1.5e+3

If we look at the output from gren-format --pre-ast, we see how "emoji" and "hexPadded" have been interpreted in a lossy way, such that gren-format can't re-write them in the same way the author originally wrote them.

 % ./gren-format/gren-format.sh --pre-ast foo.gren | grep -E "(value)"
      "value": "Repro"
    "values": [
        "value": {
          "value": {
              "value": "emoji"
              "value": {
                "value": "😀"
        "value": {
          "value": {
              "value": "letterA"
              "value": {
                "value": "A"
        "value": {
          "value": {
              "value": "carriageReturn"
              "value": {
                "value": "\r"
        "value": {
          "value": {
              "value": "hexLower"
              "value": {
                "value": {
                  "value": 255
        "value": {
          "value": {
              "value": "hexPadded"
              "value": {
                "value": {
                  "value": 255
        "value": {
          "value": {
              "value": "floatKept"
              "value": {
                "value": {
                  "value": 1500

I could work around the emoji case, by looking at the span of the string literal, and deducing whether it was written as a unicode escape sequence or not, but that's hacky.

It seems that we ought to capture the original string for Integers, Hex, CharLiteral, and StringLiteral too, and maybe PChr and PStr too?
That's a lot of strings, so maybe a better way ought to be invented?

Start by inspecting the parser and the gren-format --pre-ast output for the literal kinds named in the issue: Integer, Hex, CharLiteral, StringLiteral, PChr, and PStr. Compare the sample input with its emitted representation and determine how original spellings can be retained without losing the decoded values; done means the formatter can preserve relevant escape, hexadecimal, and numeric forms.