pk_init missing catch_panic wrapper: panic UB + stale error state

Nearly every FFI function in provekit-ffi wraps its body in catch_panic(), which:

• Calls clear_last_error() at entry
• Catches any Rust panic before it unwinds across the extern "C" boundary

But pk_init is the only function that skips this:

pub extern "C" fn pk_init() -> c_int {
    provekit_common::register_ntt();  // no catch_panic
    PKStatus::Success.into()
}

This has two major consequences:

1. Undefined behaviour on panic:

   • If register_ntt() panics, the panic may unwind through the extern "C" boundary, causing undefined behaviour in C, which can result in app (iOS/Android) crashes or silent memory corruptions with no recoverable error code.

2. Stale error state after successful init:

   • Since clear_last_error() is never called, any error set by a previous failing FFI call persists after pk_init() returns success. So, pk_get_last_error() after successful pk_init() may show a stale error message from an unrelated failure.

Search the provekit-ffi sources for pk_init and the existing catch_panic wrapper used by other FFI functions. Confirm that initialization clears the prior error state and prevents register_ntt() panics from crossing the extern "C" boundary, then run the relevant FFI tests if available.