Range mappings: Change index encoding to start at 0

[takikawa]

This implements the change discussed in the April TG4 meeting about reverting the range mapping encoding so that it will start at 0 with relative offset instead of at -1.

[jridgewell]

Nit: this means AA is valid, both marking the 0th mapping as a range.

[szuend]

One alternative that I forgot about yesterday, is that we could solve this in the grammar itself. Then you don't even need the check. We add a VlqWithoutZero or NonZeroVlq alongside Vlq that doesn't allow A.

Then the range mapping grammar is something like:

RangeMappingList :
  FirstRangeMapping
  FirstRangeMapping RangeMappingListRest

RangeMappingListRest :
  RangeMapping
  RangeMapping RangeMappingListRest

FirstRangeMapping :
  Vlq

RangeMapping :
  NonZeroVlq

Probably a lot of addition to the spec for one use-site. But maybe something to keep in mind if we ever have a need for non-zero VLQs (both signed or unsigned).

[takikawa]

I think the NonZeroVlq approach is a nice way to do it too, but for now I went with something that adds less to the core grammar.

In the latest push, I initialized the decode state's [[MappingIndex]] to NONE and then just check that a relative index of 0 can't appear if the state isn't NONE. I think this should address @jridgewell's comment about the old check not catching "AA" (thanks for pointing that out!).

[jridgewell]

Alternatively we could allow this to be negative, initialize to -1, and verify it's positive in appending the new Range Mapping Offset Record.

[nicolo-ribaudo]

I prefer ~none~, as it makes the intent clearer when reading.