rust: Close() doesn't reliably cancel listeners, causing panics

[kegsay]

Such as:

panic: Log in goroutine after TestMultiprocessInitialE2EESyncDoesntDropDeviceListUpdates has completed: [[@user-15-alice:hs1](https://matrix.to/#/@user-15-alice:hs1)]TimelineDiff change: &{ID:$sqDro8apgvTq4Gi8G-pjchrhLyQLvxC4h7jKNOjpfv8 Text:pre message Sender:[@user-16-bob:hs1](https://matrix.to/#/@user-16-bob:hs1) Targe

..despite the test calling .Close() on the client. It looks like this is a race condition as the logs show:

2024-10-16T12:04:19.458875Z  INFO TestMultiprocessInitialE2EESyncDoesntDropDeviceListUpdates: [@user-15-alice:hs1](rust) Close | rust.go:0
2024-10-16T12:04:19.458912Z DEBUG log: cancel    
2024-10-16T12:04:19.458925Z DEBUG matrix_sdk_ffi::task_handle: Cancelling the task handle | bindings/matrix-sdk-ffi/src/task_handle.rs:24
2024-10-16T12:04:19.458956Z DEBUG matrix_sdk_ffi::task_handle: Cancelling the task handle | bindings/matrix-sdk-ffi/src/task_handle.rs:24
2024-10-16T12:04:19.458655Z DEBUG matrix_sdk::sliding_sync: Done handling response | crates/matrix-sdk/src/sliding_sync/mod.rs:679 | spans: sync_once{pos="29/s159_1_0_1_8_1_76_85_0_1"}

where subsequent lines then clearly show the timeline listener being invoked, which does t.Logf which causes the panic.

Instead, we should latch closed = true when closed is called then immediately return on all callbacks if closed. It's a bit crap though.