Let an ISPDevice be driven over WebUSB

[carlossless]

Stacked on #152. Needs carlossless/hidra#9.

A device whose interface declares a vendor-specific class never appears in WebHID — Blink only surfaces interfaces the host recognises as HID — but it is reachable over WebUSB and still speaks the ISP protocol. ISPDevice now holds an ISPHandle, which is either a hidra::HidDevice or, on wasm with the new webusb feature, a hidra::webusb::HidDevice.

The protocol code is unchanged: ISPHandle dispatches the only two operations ISPDevice needs, send_feature_report and get_feature_report.

Compatibility

ISPDevice::new takes impl Into<ISPHandle> for the command handle, so existing callers pass a HidDevice unchanged. The Windows two-handle path converts explicitly — that branch is #[cfg(target_os = "windows")], so a Linux build would not have caught it; verified with cargo check --target x86_64-pc-windows-msvc.

The webusb feature is off by default and forwards to hidra/nusb; nothing changes for the CLI or the existing WebHID path.

Testing

fmt, cargo test --workspace --lib --bins, the wasm build with and without webusb, and cargo check for x86_64-pc-windows-msvc, all pass locally.

[carlossless]

Closing along with carlossless/smk-web#3 and carlossless/hidra#9.

WebUSB cannot claim the Sinowealth ISP bootloaders: they declare bInterfaceClass 0x03, which Blink blocks. Verified against a real device (258a:0036) in Chrome — it enumerates over WebUSB, but the claim is refused with "blocked because it implements a protected interface class".

So ISPHandle has no second variant worth carrying, and ISPDevice can keep holding a HidDevice directly.

Worth keeping from this: the bootloaders declare bNumEndpoints 0, so every report goes over the control pipe. That is a real property of the devices this crate drives, and the native nusb backend's control-only path is currently untested (the linux-nusb e2e uses g_hid, which always has an endpoint).