Function mark 

pub async fn mark(
    store: &dyn ObjectStore,
    prefix: &str,
    opts: MarkOpts,
) -> Result<MarkOutcome, PackchainError>

Run the mark phase: snapshot every pack on the bucket, then every chain, then write a tombstone naming the orphans.

prefix is the repository prefix without leading or trailing slashes — pass an empty string for bucket-root repositories.

Ordering

Listings run packs-first, chains-second (issue #135). The reverse order races a concurrent push that uploads a new pack between the two listings and commits its chain.json between the chain list and the pack list: the new pack would appear in the on-bucket set but not in the referenced set, producing a false-positive tombstone. Packs-first inverts the staleness: the referenced set is always at least as fresh as the on-bucket set, so a pack appearing in the snapshot is either also in the chain set (saved) or genuinely orphan at some point during the mark (correctly tombstoned, with the grace window covering in-flight pushes).

Errors

• Any chain.json that fails to parse aborts the mark with PackchainError::ParseJson / PackchainError::InvalidSha / PackchainError::UnsupportedSchemaVersion. The tombstone is not written. Operators must repair the bad chain (or remove it) before re-running.
• PackchainError::Store / PackchainError::Io for transport or local-I/O failures.

Example

use git_remote_object_store::Remote;
use git_remote_object_store::packchain::gc::{MarkOpts, mark};

let remote = Remote::connect("s3+https://bucket/repo?engine=packchain").await?;
let outcome = mark(remote.store(), remote.prefix(), MarkOpts::default()).await?;
println!(
    "{} orphan pack(s) tombstoned (run id {})",
    outcome.orphan_count, outcome.run_id,
);