- only use quiescence if both nodes signal the feature
 - reduce timeout to 1 min
 - use message type 2 for `stfu`
 - Use feature bits 34/35 for option_quiesce a per spec

It's not a prototype, it matches the official (in-progress) specification.

Use a boolean for the `initiator` field.

Add more traits to the `SpliceStatus` hierarchy, which lets us simplify
some of the pattern matching used in the `Channel` actor.

Send a `warning` and disconnect if we receive a forbidden message while
we're splicing, which lets us gracefully deal with buggy peers without
losing channels.

Being quiescent for too long is dangerous, because HTLCs may timeout
without giving the opportunity to our peer to send us a preimage.

Splicing operations shouldn't take long to complete, so we include that
in the quiescence timeout, and disconnect if the splice wasn't completed
in time. This makes sure we replay pending commands and avoid getting
the channel stuck if the interactive-tx isn't making progress.

The spec says we must not send duplicate `stfu`, but doesn't have any
requirement on the receiver. If we receive a duplicate `stfu`, it's
perfectly fine to just ignore it: the protocol will either correctly
complete or will be canceled by the quiescence timeout.

This function was called in only one place, so it's not avoiding code
duplication. It could make sense to isolate the logic, but in the case
of event handlers in the highly critical channel FSM, we usually like to
inline transition and state changes to make it easier to review the code
linearly (no need to jump back and forth between function calls).

I also renamed `isIdle` to `isQuiescent` to remove confusion.

In channel event handlers, we try to leave all the logic that updates the
current state or data directly in the event handler instead of delegating
it to helper functions, otherwise it's a bit hard to see all the state
transitions that can happen when receiving a given message. I changed the
`handleNewSplice` function to be a pure function instead that just creates
the `splice_init` message (or returns an error).

I also refactored some nits.

 - now returns a warning (and disconnects) instead of force closing

 - initiator still remains in the negotiating state until they receive stfu from the non-initiator
 - also, simplify tests for forbidden messages

 - fail htlc tests are sufficient

This commit contains almost no functional changes, we just group together
and reorder tests, remove unused code and remove the default value for
`sendInitialStfu` (which makes tests matrixes easier to read).

We were missing a test for the case where we request quiescence when we
still have pending changes to apply.

There are two scenarios to test:

- one of our outgoing HTLC times out
- one of our incoming HTLC is close to timing out and we have the preimage