refactor(cli): simplify abstractions

[psteinroe]

this PR refactors the cli crate. it currently has a bunch of issues that stem from its original implementation in biome, where file/stdin execution is the main use case.

• a CommandRunner trait that tries to abstract a few common code blocks away.
• a Dummy traversal mode which essentially means "do not traverse".
• reports are tightly coupled to execution. we also need to report output of commands that do not traverse (e.g. dblint)
• execute_mode does everything: stdin handling + traversal + reporting + exit codes. it cannot be re-used by commands that do not traverse

all of these issues cause my brain to have a hard time understanding what is going on. This PR refactors the crate to gear it more towards our use-case. I spent a few cycles designing the refactor, and at least for me it makes much more sense now.

I also added a dummy dblint command that will be implemented in follow-ups. right now, it prints an empty report.

Below is a richer summary of the issues described above.

Current Problems

1. CommandRunner Trait Forces All Commands Through Same Pipeline

pub(crate) trait CommandRunner: Sized {
    fn run(&mut self, session: CliSession, cli_options: &CliOptions) -> Result<()> {
        // Forces: config loading → VCS setup → file traversal → reporting
        let (execution, paths) = self.configure_workspace(fs, console, workspace, cli_options)?;
        execute_mode(execution, session, cli_options, paths)
    }
}

Issues:

• Only Check command uses this trait
• Commands like version, clean, init are ad-hoc functions
• No way for commands to skip parts they don't need (e.g., dblint needs config but not traversal)
• Forces unnecessary complexity on simple commands

2. TraversalMode::Dummy Exists For Wrong Reasons

pub enum TraversalMode {
    Dummy,  // Only exists because commands are forced through traversal
    Check { stdin: Option<Stdin>, vcs_targeted: VcsTargeted },
}

Commands that don't need traversal shouldn't have a "dummy" mode - they just shouldn't traverse.

3. Execution Struct Conflates Concerns

pub struct Execution {
    report_mode: ReportMode,      // How to report (Terminal/GitHub/GitLab)
    traversal_mode: TraversalMode, // How to process files
    max_diagnostics: u32,
}

Problem: Bundles processing config with reporting config. Commands that don't traverse (like dblint) still need reporting but don't need traversal config.

4. execute_mode() is Monolithic

pub fn execute_mode(execution: Execution, session: CliSession, ...) -> Result<()> {
    // Does: stdin handling + traversal + reporting + exit codes
    // Can't be reused by commands that don't traverse
}

5. Scattered and Inconsistent Structure

• commands/ has mix of trait impls and functions
• execute/ has traversal logic tightly coupled to reporting
• No clear separation of workspace setup, execution, and reporting
• Commands repeat boilerplate for config loading, setup, reporting

[juleswritescode]

much simpler! love it.

[juleswritescode]

these seem like useful comments?

[juleswritescode]

this is clearly a peak MVP

[juleswritescode]

these seem to be dead, do we need them?
if not, let's remove the unnecessary complexity 👍🏻

[juleswritescode]

I love that the trait was called "Visitor" before, with a command specifying that it would write

[juleswritescode]

sanity check: is the comment regarding "skipped diagnostics" counter useful?

[juleswritescode]

this is still confusing to me:

1. apparently file is never read – should we delete it?
2. if we want to keep it, how can a WorkspaceFile not be backed by a real fs entry? In case it's a temporary buffer I suppose – but let's add some clarification in a comment 👍🏻

[psteinroe]

its for stdin execution - which we currently do not have a real use for since we never write in our commands. this will become used once we have autofix.