Draco-compressed version of VirtualCity might be invalid

[javagl]

The Draco-compressed version of the VirtualCity may be invalid:

• The mesh with index 15 refers to accessor 202 as its indices
• The accessor with index 202 contains a max: [ 228 ]
• The mesh refers to the accessor with index 203 as its NORMAL
• The accessor with index 203 (just below the indices) claims a count of 140 - meaning that some indices will be out of bounds for that

For comparison: In the glTF version (default, uncompressed):

• The mesh with index 15 refers to accessor 88 as its indices
• The accessor with index 88 contains the same max: [ 228 ]
• The mesh refers to the accessor with index 89 as its NORMAL
• But the accessor with index 89 (just below the indices) has a count of 229 - just right for the maximum index of 228

Maybe I'm overlooking something in the Draco spec - something that claims that the count may be that of the compressed/reordered data or something. The Conformance section claims that a loader must ignore the data of the accessor, and use the Draco-decoded data instead - but I think that this does not imply that the count of these accessors may be wrong.

There probably aren't many people who can quickly give a 👍 or 👎 on whether this is really invalid. But I hope that it's OK to tag @lexaknyazev here...

---

_{EDIT: Mistyped "accessor" as "buffer view" in some places...}