fix(client): release in-flight commands when the connection dies

[route]

Refs #470.

The problem

Client#send_message waits for the response and only afterwards asks whether the connection is still alive:

data = pending.value!(timeout || protocol_timeout)
@pendings.delete(message[:id])

raise DeadBrowserError if data.nil? && @ws.messages.closed?
raise TimeoutError unless data

So when the browser dies, nothing wakes the commands already in flight, and nothing stops new ones from being sent into a closed socket either. Each pays the full :protocol_timeout before raising. In a test suite that means one dead browser turns into hundreds of failures at five seconds apiece, which is a good part of what makes #470's collapse so slow and so hard to read.

The fix

The dispatch loop releases the pending commands as soon as the message queue closes, and a command sent after that raises DeadBrowserError immediately:

raise DeadBrowserError if @ws.messages.closed?

release_pendings   # after the loop breaks on a closed queue

No new error classes and no change to what callers see, only when they see it: DeadBrowserError was already what these paths raised, just five seconds later.

Verification

• New example in spec/browser_spec.rb #crash: crashing the browser raises DeadBrowserError in under 2s, and so does the next command against the dead browser. Before the fix the first assertion fails at the full 5s timeout.
• Full suite: 616 examples, 0 failures.