refactor(error): per-transport TransportError and generic ceremony Error

[AlfioEmanueleFresta]

Today every transport failure collapses into one catch-all transport error, and the underlying cause from the OS, Bluetooth, USB, or tunnel layer is discarded. That makes diagnostics weak and leaves the error surface unsettled ahead of a 1.0 freeze.

This reworks the error model so the transport error type is known at compile time and the cause is preserved.

Changes

• The Channel trait gains an associated TransportError type. Each transport (HID, BLE, NFC, hybrid) has its own concrete error enum that carries the native cause instead of flattening it away.
• The WebAuthn ceremony error becomes generic over the channel's error type. Running a ceremony on a concrete channel yields a concrete, matchable error with no boxing and no downcasting.
• The CTAP protocol error variant stays transport independent, so code that matches on it is unaffected.
• Request preparation keeps its own validation error types. A preparation error is a distinct phase and is not representable as a ceremony error.

The ceremony error:

pub enum Error<TE> {
    Ctap(CtapError),
    Transport(TE),
    Platform(PlatformError),
}

Closes #130.