feat(rivetkit): re-export serverless_http and default absent actor input

[abcxff]

No description provided.

[abcxff]

Stack for rivet-dev/rivet

• #5394 refactor(container-runner): rewrite on rivetkit actor runtim …
• #5395 feat(rivetkit): re-export serverless_http and default absent … 👈
• #5392 docs(deploy): add container runner page
• #5393 ci(publish): build and publish container runner release arti …
• #5391 feat(container-runner): add container runner crate and e2e h …
• main

Get stack: forklift get 5395
Push local edits: forklift submit
Merge when ready: forklift merge 5395

_{change kknzrylw}

[claude]

Review

Reviewed the diff (Cargo.toml, actor.rs, lib.rs, start.rs).

Summary of the change:

• Registers rivetkit in [workspace.dependencies] so other in-repo crates can depend on it via workspace = true.
• Re-exports rivetkit_core::serverless_http from the rivetkit crate.
• Adds a Default bound to Actor::Input and switches run_actor to call input.decode_or_default() instead of input.decode(), so a missing input now falls back to Default::default() instead of erroring with MissingInput. The comment correctly explains this mirrors rivetkit-typescript, where createState just receives undefined.
• Updates test fixtures (LifecycleInput, UnitInput) to derive Default and fixes several test-helper Start/Snapshot struct literals that were missing the is_new field (pre-existing from an earlier stacked change) so the crate compiles again.

Findings:

1. Test coverage gap for the actual behavior change. The new default-on-missing-input path is only exercised at the Input::decode_or_default() unit level (input_decode_or_default_uses_default_when_missing). There's no run_actor-level test (like run_actor_missing_unit_input_creates_state, but for a non-unit Default input) that proves create_state actually receives the defaulted value end-to-end when no input bytes are sent. The DefaultActor/DefaultInput fixtures added for the unit test don't implement create_state, so they can't currently be reused for that. Worth adding a run_actor-level integration test to lock in the new lifecycle behavior, since this is the actual production code path being changed.

2. Behavioral/breaking-change note (minor). Adding Default as a supertrait bound on Actor::Input changes the failure mode for any actor whose input is logically required: previously, a missing input would surface as an explicit ActorRuntime::MissingInput error; now it silently proceeds with Default::default() for any type where that's meaningful (not just unit-like types). This is presumably intentional (matches TS parity, and rivetkit (Rust) is called out as a preview API in CLAUDE.md), but it's a real semantic change for anyone relying on the old "missing input fails loudly" behavior, and it's also a breaking API change for downstream implementors whose Input type doesn't derive Default. Might be worth a one-line callout in the PR description / changelog.

3. Nit: unconditional re-export of a feature-gated module. rivetkit_core::serverless_http is behind #[cfg(feature = "native-runtime")] in rivetkit-core. The new pub use rivetkit_core::serverless_http; in rivetkit/src/lib.rs is unconditional. Today this is safe because rivetkit-core's default features (and thus rivetkit's dependency on it) include native-runtime, and rivetkit isn't part of the wasm build gate (scripts/cargo/check-rivetkit-core-wasm.sh only checks -p rivetkit-core). Just flagging so this doesn't bite later if rivetkit ever grows a wasm-targeted build.

No security or performance concerns — this is internal SDK/lifecycle wiring with no new trust-boundary-crossing code. Code style (comments, hard tabs, derive placement) matches CLAUDE.md conventions.