release: commitward 0.3.0

[lyubomir-bozhinov]

Version bump only — Cargo.toml + Cargo.lock, matching 7ed509a (release: commitward 0.2.1). Tagging and publishing stay human steps.

Why 0.3.0 and not 0.2.3

No signature moved, but compile() behaves differently for every caller (#9):

pub fn compile(cps: Vec<Checkpoint>) -> Result<Vec<CompiledCheckpoint>, CheckpointError> {
    merge(cps, anchor_checkpoints())   // ← anchor last, so an on-disk entry cannot shadow it
        .into_iter().map(compile_one).collect()
}

A caller that compiles N checkpoints now gets N+1 back. Every compiled registry — including an empty one — carries anchor-gate-integrity, which watches the gate's own registry and hook files and is compiled into the binary, so no YAML edit removes it.

That is the point rather than a side effect: the registry could not be the sole protector of the registry, so one checkpoint now lives outside it.

What consumers will feel

A commit touching a registry or the commit-msg hook now always fires at least one checkpoint — including the commit that first adopts a registry. Acknowledge it with a HITL-ACK line like any other fire; it fires, it does not block.

The NF3 "no checkpoints supplied" warning still fires when the caller supplied nothing, with its claim narrowed to what is now true — the compiled set is never empty, so the warning asks about the caller's input instead.

Additive in the same release: anchor_checkpoints() is public, so a consumer can inspect the floor rather than trust it.

Verification

cargo test --locked 40 passed / 0 failed · lock refreshed via cargo check.