[forhappy]

The JavaScript SDK and bridge were difficult to read, and the SDK README lacked complete usage examples. This makes the public JavaScript/TypeScript API, shared facade, Rust adapter, and Node worker easier to follow while preserving their behavior and public APIs.

Changes

• Expand condensed SDK implementation and type declarations; document requests/responses, storage, alarms, bindings, errors, and invocation lifetime with concrete examples.
• Expand bridge protocol payloads and worker control flow. Explain cancellation, activation ownership, byte budgets, mutation outcomes, continuation identity, and process shutdown.
• Align Cellar-owned Cargo and npm package versions and lockfile entries to 0.1.0, preserving third-party versions.
• Add a repeatable V8/Node TypeScript benchmark with HTTP throughput, durable write TPS, latency distributions, resource samples, contention checks, and cold-storage reconciliation. Include the measurement report and portable evidence.

Validation

On updated main, using macOS arm64, Node 24.13.1, and Rust 1.98.1:

• 21 Rust tests passed with both embed,node features: context isolation, cancellation, mutation refusal, watchdog/process cleanup, IPC framing, npm/native-addon compatibility, and storage recovery after restart.
• Four JavaScript build tests passed; both example applications and benchmark TypeScript compile.
• Affected-crate Clippy with warnings denied, repository format/layout, native dependency policy, documentation references, and diff whitespace checks passed. Existing vendor warnings remain.
• The shared facade and Node worker produce identical normalized executable JavaScript before and after the readability edits.
• Earlier SDK validation compiled all seven README TypeScript examples and ran the documented counter against V8 and Node, including fresh-node recovery.

Performance evidence and limits

The included campaign was measured at the explicitly recorded earlier source revision, before the version/readability changes; it is not a new measurement of this PR head. On the shared development machine, median 32-client 1 KiB echo throughput was 14,154 req/s for V8 and 11,088 for Node; sharded durable writes were 459 and 451 TPS. All 41,344 acknowledged increments, including warmup, reconciled after cold recovery.

The report preserves hot-cell conflicts and the V8 64 KiB/32-client failures, including one confirmed execution-budget failure and another dispatch failure whose cause was not established. This PR does not fix those runtime limitations. Results use a local SQLite development bucket and closed-loop traffic; they do not establish production fleet capacity.