feat: Add a content field and element to raw string literals.

[Natr1x]

This makes it possible to use language injection for this content.
Which is useful since this is something that raw string literals are often used for.

A query like the following could for example be used to provide sql highlighting when using sqlx:

; extends
((macro_invocation
   macro: (scoped_identifier
            path: (identifier) @macropath (#eq? @macropath "sqlx")
            name: (identifier) @macroname)
   (token_tree
     (raw_string_literal
       content: (_) @injection.content)))
 (#set! injection.language "sql")
)

Note: I have not ran tree-sitter-cli generate in order to keep the commit smaller for review purposes.

If you are interested in merging this but want me to change or add something (or run that code generation) then I am willing to do so.

[amaanq]

A struct for one field is a bit overkill, let's just allocate a uint32_t instead (I'd rather not use unsigned)

[Natr1x]

Are you sure about this? Personally I feel keeping it in a struct makes it explicitly clear that this is the only state we keep track of.

[amaanq]

alright yeah, if we do need to add more fields later on for whatever reason it'd be easier. Can we declare it as a typedef, use uint32_t, and call it just Scanner though?

typedef struct {
    uint32_t opening_hash_count;
} Scanner;

[Natr1x]

I am not a big fan of anonymous structs, would f6caf4a work as a compromise?

If not I am okay with removing the name. It's mostly for the sake of things like lsp hover information.

[mkatychev]

@Natr1x what else needs to get done for this PR to move out of draft. This would be extremely useful for sql injections and would remove #offset! guesswork.

[amaanq]

reworked it in #220