Bug in SemanticTokensDocument.GetSemanticTokenEdits

This bit of code in GetSemanticTokenEdits (lines 162 to 178) throws an ArgumentOutOfRangeException in the call to ImmutableArray.Create when you delete code in a document (but maybe in other cases too):

while (startIndex < dataLength && startIndex < prevDataLength && prevData[startIndex] ==
    Data[startIndex])
{
    startIndex++;
}

if (startIndex < dataLength && startIndex < prevDataLength)
{
    // Find end index
    var endIndex = 0;
    while (endIndex < dataLength && endIndex < prevDataLength &&
           prevData[prevDataLength - 1 - endIndex] == Data[dataLength - 1 - endIndex])
    {
        endIndex++;
    }

    var newData = ImmutableArray.Create(Data, startIndex, dataLength - endIndex - startIndex);

It looks like the intent here is to find both the longest common prefix and the longest common suffix between prevData and Data. Then it will return a SematicTokensEdit with a DeleteCount = prevDataLength - endIndex - startIndex to delete the tokens in the middle, and Data = newData to insert the replacement tokens.

Logically:

prevData = [... common prefix ...] [... stuff to delete ...] [... common suffix ...]
    Data = [... common prefix ...] [... stuff to insert ...] [... common suffix ...]

Where "stuff to insert" can have a different length than "stuff to delete", including a length of zero.

So it's trying to diff them like this:

prevData = [a a a b b b c c c]
    Data = [a a a       c c c] // extra spaces added for alignment, Data.Length = 6
                  ^ ^ ^ snip!

startIndex = 3
endIndex = 3

newData = Create(Data, 3, 6 - 3 - 3) // length = 0, this is fine
DeleteCount = 9 - 3 - 3 // = 3, all good

But this isn't reliable. The common prefix can overlap with the common suffix, just by chance, as in this example:

prevData = [a a a a b a a a a]
    Data = [a a a       a a a]
                  ^ ^ ^ snip!

startIndex = 4
endIndex = 4

newData = Create(Data, 3, 6 - 4 - 4) // length = -2, oops!!!!
DeleteCount = 9 - 3 - 3 // = 3

The search for endIndex needs to be limited to not eat into the common prefix. In this example, we should get startIndex = 4, endIndex = 2, DeleteCount = 3, newData.Length = 0.

Start by reading SemanticTokensDocument.GetSemanticTokenEdits around lines 162-178 and trace the prefix and suffix scans for the deletion example. Reproduce the overlapping-prefix case, then verify the resulting SemanticTokensEdit has valid replacement and delete lengths without an ArgumentOutOfRangeException.