Report worker termination to channel tasks

[paberr]

A channel task could not survive the death of its worker:

• ChannelTask::result() ended in .expect("WebWorker result sender dropped"), so a
  terminated worker turned a pending result() into a panic. Dropping a WebWorker
  is enough to trigger this, since Drop already calls worker.terminate().
• ChannelTask::recv() could never return None. Channel::from calls
  callback_handle.forget(), so the mpsc sender is leaked and the receiver never
  closes. On a dead worker recv() blocks forever. The doc comment claiming it returns
  None "if the worker has finished and closed the channel" described a path that did
  not exist.

Both are reachable today without any new API, so worker death was effectively
unobservable: one branch panicked, the other hung.

Changes

• result() returns Result<R, TaskError>; TaskError::WorkerTerminated reports that
  the worker went away before returning a result.
• recv() / recv_bytes() return None once the worker is terminated. Messages that
  already arrived are still handed out first, so nothing in flight is lost.
• WebWorker::terminate() and is_terminated() are now public. terminate() closes the
  port, terminates the worker and drops the pending task senders; Drop just calls it.
• Calling run/run_channel on a terminated worker fails fast with a clear message
  instead of posting into a dead port and hanging forever.

Terminating is the only way to stop work that does not cooperate — a computation without
a cancellation point, or a task that spawned background work of its own. Cooperative
cancellation over the channel stays the recommended path for everything else, since
terminating discards a warm worker.

Scope

This is the subset of #22 that is a bug fix. It deliberately leaves out
ChannelTask::terminate() and the pool's exclusive worker leases: killing a worker from
a task handle is only safe when the task owns that worker, and making every pool channel
task own one blocks ordinary pool.run() calls for as long as a channel task lives.
The use case behind #22 is served here by running such a task on its own WebWorker.

pool.run_channel() is unchanged and keeps sharing workers.

Breaking changes

• ChannelTask::result() returns Result<R, TaskError>.
• ChannelTask::new() is now pub(crate) (it was pub but #[doc(hidden)], and
  callers outside the crate could not construct its arguments).

No version bump in this PR — happy to add one if you want it in the same commit.

Closes nothing on its own; #22 remains open for the termination API discussion.

🤖 Generated with Claude Code

Unrelated: clippy fix

The second commit silences clippy::clone_on_copy on the getters that
#[wasm_bindgen(getter_with_clone)] generates for WorkerPoolOptions. This is a
pre-existing failure on main that a recent nightly clippy started reporting, not
something this PR introduces — the only other change here to src/pool/mod.rs is a
doc-comment line. It is in its own commit so it can be split off and landed on main
separately if you prefer; main needs it either way.

[paberr]

Also, thanks a lot to @Roba1993 for bringing up the issue and opening the initial PR! 👏
@Roba1993 If you want to have a look and see if this is useful to you, I'd appreciate it!