chore: revamp attributes

[asterite]

Description

Follow-up to #6414

Problem

Currently attributes, even "meta" ones (non "tag" ones) were parsed by just consuming all characters until hitting "]". That worked but:

• it didn't allow to use arrays in arguments
• I think in general we wanted to make meta attributes always having the func(arg1, .., argN) or just func syntax, and kind of enforcing that

Summary

This PR changes how attributes are parsed and represented. Meta attributes must now always have a function syntax call. The rest of the attributes (the built-in ones) are parsed in a more strict way, generally leading to better error messages.

Some benefits of this PR are:

• We no longer need to parse an attribute to see if it's a function call
• Any kind of expression is allowed as a meta attribute argument
• Comments are now allowed inside attributes
• Attributes are formatted better
• LSP will offer better completions: because now the meta arguments are expressions, if you type #[derive(E it will offer Eq as a suggestion.

Additional Context

I had to comment out a #[deprecated = "..."] attribute for two reasons:

1. That syntax was wrong (the correct one is #[deprecated("...")]
2. If we change it to use the correct syntax, we get deprecation warnings in the standard library... which is probably not something we want. I tried to defined an alternative method that isn't marked as deprecated and use it in the standard library (marked as pub(crate)) but some tests rely on that new method and I didn't know how to make it work otherwise.

Of course, let me know if the direction this PR is going isn't the desired one!

Documentation

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[jfecher]

Of course, let me know if the direction this PR is going isn't the desired one!

I haven't seen the line by line changes yet but I'd rather keep the ability to call the functions without an explicit function call syntax. E.g. still allowing #[aztec_private] and not requiring #[aztec_private()]. I'd think we should still be able to support this so even while parsing these so I'm curious why it was removed?

[asterite]

Oh, it should still be allowed. If not I might have made a mistake.

[jfecher]

Ah, ok. I just assumed from the description it sounded like a function call form was required. Should be fine then

[asterite]

Ah, makes sense. I updated the description.

[asterite]

A nice side-effect of this PR is that because function attribute arguments are parsed as a regular call, their location is the one in the source file. Before this PR their location is relative to the start of the attribute, leading to these errors:

Of course that can be fixed by shifting all locations, but this PR makes it work out of the box.

[jfecher]

Looks great, thank you!