Scopes: Update encoding to 'tags' approach

[szuend]

This PR changes the encoding of the "Scopes" proposal to the "Tag" approach presented in the TG4 meeting of 2024-01-30 (see slides.

The information contained is exactly the same, we just use a different encoding scheme to be future-proof.

The PR includes some changes w.r.t. to the slides. These are of course up for discussion:

• The PR stores binding expressions in a separate JSON field "expressions" rather than piggy-backing on "names". This was discussed in some forums. I'm not 100% sure about this as some strings may end up twice in the source map this way. Especially if no variable renaming happened.
• The slides didn't contain a way how to connect original scope trees to source file indices. The PR includes a new (optional) original_scope_source_index item that can be used to connect a root original scope tree with a specific source file. A second way is to include exactly as many root original scope trees as there are source files, so the connection is implicit via indices. The generator can choose the approach.
• The PR moves FLAGS in original_scope_start to the beginning to. be more consistent with generated_range_start.
• DEFINITION in generated_range_start is not an item index, but rather the nth original_scope_start item to further reduce the magnitude of the value.
• The PR splits bindings into 2 separate items: The general case is a simple list of expressions that map 1:1 to the variables list. I.e. the variables use the same expression throughout the full generated range. The subrange mechanism was split into its own item: It contains a variable index, so it can be applied to some variables. This allows us to encode expressions relative to each other and we no longer have the shenanigans with -1 meaning unavailable and < -1 means "number of subranges".

The wording is still ambiguous in some cases, but should be (mostly) enough for the proposal doc.