Add function annotations and fix Clang error

[j-walther]

Hi!

This PR changes a few things to accomodate the following new annotations:

• function-arg-prefix[arg]
• function-arg-ident-prefix[arg]
• function-prefix
• function-postfix
• function-ident-prefix

Annotations

I've overhauled the list parsing to allow for some special characters to be escaped - as I've noted in the docs.

In addition to that I've introduced a new data type for annotations called Dict. It's a mapping of string indices to annotation values.
This was necessary to accomodate the function-arg-prefix[arg] and function-arg-ident-prefix[arg] annotations so that in a function like

fn foo(bar: *const usize) {}

annotations to arguments can simply be created by adding cbindgen:function-arg-prefix[bar]. I much prefer this over indexing by number since that might inadvertently add the annotation to another argument if they're ever shifted around.

As a part of these changes I've added the CFuncArg struct to replace the previous tuple. Initially this was because I placed the ident_prefix there but it turned out to be easier to just reuse the one on CDecl.

These annotations allow specifying SAL annotations, Compiler-specific attributes (e.g. Clang) and macros in general.

Clang errors

The body example added an inline void foo() to body.compat.c which caused a compiler error in conjunction with -Wnon-c-typedef-for-linkage since the generated structs were untagged. This effectively broke the automated test suite for me.

I've changed it so that if cbindgen is generating C++ with C compatibility and the user adds anything using [export.body] or [export.pre_body] cbindgen will enforce tags for those structs.

Function affixes

The remaining annotations provide similar benefits but for functions.

Outlook

This PR lays the foundation for solving #982. For more advanced use-cases the annotations defined in this PR should still be available but hidden behind an abstraction layer for e.g. NonNull parameters.