A live screen that ends tells the app, and the app renders it in the branch it is not in

[beardthelion]

The server tells a viewer when its live screen ends. The app takes that message, stores it, and then
renders it only in the branch it is not in.

LiveScreen forwards every {type: "error"} frame to onProblem
(app/src/components/computer/live-screen.tsx:79-80), and ComputerView sets that into problem
(computer-view.tsx:224, :646). The only thing that draws problem is NothingToSee
(:153-158), and the two are alternatives of the same ternary at :642:

) : showLiveScreen ? (
  <LiveScreen ... onProblem={setProblem} />
) : (
  <NothingToSee ... problem={problem} />
)

So while the live screen is mounted, which is exactly when a person is watching, the message lands in
state that nothing on screen is reading. showLiveScreen is !settled && (showScreen || driving)
(:432), so it stays true while somebody is driving, which is when they are most likely to be told
something.

What a person sees instead is the last frame that arrived, indefinitely. The canvas is not cleared,
so a screen that has ended looks exactly like a screen where nothing is happening. Their input goes
nowhere and they are not told why.

Three cases reach this today, and there is a fourth once #190 is fixed:

• Another window took the screen. A second /stream replaces the first, and the replaced socket is
  told rather than closed, because it belongs to a client that may still be using it.
• The socket could not be reached at all (live-screen.tsx:120).
• The screen could not be started (the open handler's failure path).
• After The live screen is tracked per Bot rather than per socket, and four failures follow from it #190: the computer stopped, was reset, or its browser was closed by the cap or the idle
  sweep. Each of those now sends the watching socket a message saying so.

I have a branch for the #190 half, which is all server-side and leaves this untouched. Worth saying
plainly that the server work does not fix what a person sees: the message arrives at the socket and
stops there.

Not a security or data problem. It is somebody watching a still image and not knowing the screen is
over.