[dpep]

Description

When the local console hits EOF, UI_LocalConsole#readline sends quit, which asks "Really quit?" via UI_LocalConsole#ask. ask reads $stdin, but with Reline the prompt reads STDIN. So if STDIN is exhausted and $stdin has been replaced by something unreadable, the confirmation raises IOError. process_command reports it as [REPL ERROR] and retries, the next readline hits EOF again, and the command loop spins forever printing:

Really quit? [Y/n] [REPL ERROR] #<IOError: not opened for reading>

Repro (debug 1.11.1 and master; in CI this hangs and fills the log):

# repro.rb
require 'debug'
require 'stringio'
$stdin = StringIO.new("input").tap(&:close)  # e.g. code under test swapping $stdin
debugger

$ ruby repro.rb < /dev/null

Fix: UI_LocalConsole#ask answers an unreadable $stdin the way it answers EOF, with the default, so the session quits as it already does on EOF. Remote sessions use UI_ServerBase#ask and are unchanged.

Test: a new case in test/console/quit_test.rb. The console tests run with RUBY_DEBUG_NO_RELINE, where readline also reads $stdin, so the test installs a fake $stdin that returns EOF and then raises; the extra type steps only drive the harness's queued-command requirement. It fails with the [REPL ERROR] loop without the fix and passes with it. rake test_console passes locally (Ruby 4.0.6), and quit_test.rb passes on 3.4.10.

A deeper alternative would be to have ask read through the same input as readline, so both always see the same stream. That changes what users see at the confirmation prompt, so I kept this to the minimal fix — happy to go that way instead if you prefer.