Support multiline and overlapping semantic tokens

There are client capabilities for if the client supports these. But if they don't, we can still support them gracefully I think. The conversion functions can take the client capabilities and then implement a workaround if the capabilities aren't supported:
- For overlapping tokens, split the overlapped token so into the part before the overlapping token and the part after the overlapping token, so you get something like: `["hello ${a} world"][a]` -> `["hello ${][a][} world"]`
- For multiline tokens, split them at the line breaks into one token per line.

Start by locating the conversion functions that produce semantic tokens and the client-capability data they receive. Check how overlapping and multiline tokens are represented, then verify that unsupported capabilities are handled by splitting overlaps and line breaks into valid per-line tokens.

Written by the indexing model from the issue text.