feat: add owned typed arrays

[GrapeBaBa]

Summary

• add js.OwnedTypedArray(Element, array_type) and concrete Owned*Array aliases
• transfer allocator-owned element storage to an external ArrayBuffer without copying
• support owned typed arrays as return values from DSL functions and class methods
• make ownership transfer transactional: pre-transfer failures preserve the Zig owner; successful or potentially transferred values leave the source empty
• validate each Element against its N-API TypedArray kind at compile time
• keep TypedArray.fromExternal unchanged from main
• propagate NoExternalBuffersAllowed unchanged; no copy fallback

API

pub fn serialize() !js.OwnedUint8Array {
    const allocator = js.allocator();
    const data = try allocator.alloc(u8, 32);
    // Fill data...
    return js.OwnedUint8Array.fromOwnedSlice(allocator, data);
}

fromOwnedSlice takes an existing allocator-owned mutable slice without copying. fromSlice creates an owned copy when that is what the caller needs.

intoValue takes *OwnedTypedArray. Once ownership transfers, it empties the source, so a normal deferred deinit is safe. The allocator must remain valid until the ArrayBuffer finalizer runs.

Ownership boundary

• zero length: create a normal zero-length V8 ArrayBuffer and empty the source only after TypedArray creation succeeds
• owner heap allocation failure: leave the allocation owned by the caller
• NoExternalBuffersAllowed, PendingException, or CannotRunJS: Node returns before finalizer registration, so restore ownership to the caller and propagate the original error
• other external ArrayBuffer errors: leave the source empty because Node may already have installed or invoked the finalizer
• typed-array view creation failure or success: leave the source empty; the external ArrayBuffer finalizer owns the allocation
• DSL return conversion: defer deinit on the local owner, which frees restored/untransferred data and is a no-op after transfer

The heap finalizer context is the OwnedTypedArray itself, matching the boxed-owner model rather than duplicating ownership fields in a second context type.

Node implementation reference for the ambiguous failure boundary:

• https://github.com/nodejs/node/blob/main/src/js_native_api_v8.cc#L2867-L2882

Verification

• zig build test:zapi
• zig build
• pnpm test:js — 119/119
• pnpm lint:js
• zig fmt --check src/js/typed_arrays.zig src/js/wrap_function.zig
• git diff --check origin/main
• independent Zig ownership audit — PASS, 0 MUST / 0 SHOULD

Spec tests were intentionally not run.

The repository-wide zig fmt --check src still reports the pre-existing formatting issue in src/create_callback.zig; this PR does not modify that file.

Coverage limitation

The JS tests exercise non-empty and zero-length OwnedUint8Array values through the real Node addon boundary. They do not deterministically force GC to count finalizer invocations or inject NoExternalBuffersAllowed, pending/cannot-run, generic external-creation, or typed-array-view creation failures. No production test seam or compatibility branch was added for those cases.