Fix byte count in NodeAnyTypedArray.withUnsafeMutableRawBytes

[liautaud]

Summary

NodeAnyTypedArray.withUnsafeMutableRawBytes currently passes the length value returned by napi_get_typedarray_info straight into UnsafeMutableRawBufferPointer(start:count:). Per the Node-API docs, that value is the element count of the typed array, not its byte count. The two only coincide for Int8 / UInt8 / Uint8Clamped arrays.

For every other typed-array kind, the returned raw buffer is silently truncated. This is particularly visible through NodeTypedArray<Element>.withUnsafeMutableBytes, which calls bindMemory(to: Element.self) on the returned raw buffer: the resulting UnsafeMutableBufferPointer<Element> reports a count divided by MemoryLayout<Element>.size, so callers only see a fraction of the actual elements (and any reads/writes beyond that fraction are silently skipped).

Fix

• Query the typed-array kind from napi_get_typedarray_info (instead of passing nil for the type out-parameter).
• Multiply the returned element count by the per-element byte size to get the real byte count.
• Hand back an UnsafeMutableRawBufferPointer that covers the entire typed array.

A new byteSize property on NodeTypedArrayKind centralises the element-size mapping and mirrors the existing raw property.

Repro

let array = try NodeTypedArray<Float>(for: buffer, count: 1024)
try array.withUnsafeMutableBytes { buf in
    print(buf.count) // before: 256, after: 1024
}

Notes

• No behaviour change for Int8 / UInt8 / Uint8Clamped arrays.
• The pre-existing // TODO: Do we need to advance data by offset/a multiple of offset? is removed; napi_get_typedarray_info already returns data adjusted by the byte offset, per the Node-API spec.

[liautaud]

(cc. @kabiroberai)