Serialize JSON report references as HTML color codes

[DJMcNab]

Follow-up from #1328

The new formatting looks like:

...
{
  "x": 1,
  "y": 18,
  "reference": "#ece5f2",
  "actual": "#bfffbf",
  "difference": [
    -45,
    26,
    -51,
    0
  ]
},
{
  "x": 2,
  "y": 1,
  "reference": "#d9cce5",
  "actual": "#7fff7f",
  "difference": [
    -90,
    51,
    -102,
    0
  ]
},
...

I've also tweaked the in-repo settings of vscode to show the colour picker in these diff files:

[grebmeg]

Thanks for the follow-up! The implementation looks easier than I thought it would be. I’d keep the numeric values alongside the hex values, since I have a feeling that will make debugging easier, especially if AI is involved. I don’t have strong evidence for that, just an assumption.

Also, what do you think about using reference and actual names of the same length to make eye-scanning easier, for example ref/act/diff or maybe source/target?

[DJMcNab]

On keeping the full decimal values around, personally I'm not convinced that adds any value. The motivation for this PR was to remove the verbosity that keeping the numbers fully required. I also think that an LLM is likely to be more "familiar" with hex formatted colours than lists of u8 components. Unless you have evidence that it's needed, I'd much prefer to leave it out.

On matching the lengths, that's a very good idea. It's important to keep the old and the new distinct/make it clear which is which, so I'll workshop the names tomorrow. reference and observed, which spring to mind, are a bit too "cute". Maybe target and actual?

[grebmeg]

On keeping the full decimal values around, personally I'm not convinced that adds any value. The motivation for this PR was to remove the verbosity that keeping the numbers fully required. I also think that an LLM is likely to be more "familiar" with hex formatted colours than lists of u8 components. Unless you have evidence that it's needed, I'd much prefer to leave it out.

Yeah, I don’t have it, so let’s leave it as it is. Thank you!

On matching the lengths, that's a very good idea. It's important to keep the old and the new distinct/make it clear which is which, so I'll workshop the names tomorrow. reference and observed, which spring to mind, are a bit too "cute". Maybe target and actual?

I like target/actual and agree that clearer naming here would help. Some other options I like: gold/test, source/actual, and expect/actual.