fix(client): keep Ferrum's threads from taking the host process down

[route]

Refs #470.

The problem

Utils::Thread.spawn defaulted to abort_on_exception: true, and the two transport threads used that default. The client dispatch thread has no rescue at all, and the websocket reader rescues only EOFError, ECONNRESET, EPIPE, IOError. Anything else is re-raised in the main thread wherever it happens to be.

Forcing one on main, an error the reader doesn't expect:

b = Ferrum::Browser.new
page = b.create_page
driver = b.client.instance_variable_get(:@ws).instance_variable_get(:@driver)
def driver.parse(_data) = raise(Errno::ETIMEDOUT)

page.go_to("data:text/html,<h1>hi</h1>")   # Errno::ETIMEDOUT, out of a navigation

In a suite that's a failure attributed to whatever call was unlucky enough to be running. In a server it's process death, with the caller's ensure never getting to browser.quit.

The change

• Utils::Thread.spawn defaults to abort_on_exception: false. Every call site already passed false except these two, so the explicit arguments are gone and the intent lives in one place. The deliberate main-thread raise for JS errors is untouched: it uses Thread.main.raise directly.
• Both transport threads rescue and report to stderr, and the websocket reader closes @messages in an ensure rather than only on expected disconnects.
• Either thread dying now means the connection is dead, so callers get Ferrum::DeadBrowserError rather than an error from a place they never called.

The snippet above becomes:

Ferrum: websocket reader stopped, Errno::ETIMEDOUT: Operation timed out
  ...
Ferrum::DeadBrowserError

Verification

• New example in spec/client_spec.rb: an unexpected error in the reader is reported and the next command raises DeadBrowserError. Without the fix the example fails with expected Ferrum::DeadBrowserError, got #<Errno::ETIMEDOUT>.
• Full suite: 616 examples, 0 failures.

Overlaps #630 slightly: there the pendings are released when the queue closes, so with both merged the DeadBrowserError here arrives immediately rather than after :protocol_timeout. Trivial conflict in Client#start if they land in the same window.