Gracefully handle non-existing failing test given to bisect

[zarifmahfuz]

Since we started periodically reproducing state leaks, I saw some cases where the failing test does not exist anymore. In such cases, the bisect command thinks that the failing test independently fails because the run_tests_in_fork(queue.failing_test) returns false -

--- Testing the failing test in isolation
/tmp/bundle/ruby/3.4.0+0/gems/ci-queue-0.52.0/lib/ci/queue/static.rb:91:in 'Hash#fetch': key not found: "Sales::ApiOrderToOrderServiceTest#test_an_order_... (KeyError)
--
  | Did you mean?  "Sales::OrderEditing::LineItemBuilderTest#test_some_other_test"
  | from /tmp/bundle/ruby/3.4.0+0/gems/ci-queue-0.52.0/lib/ci/queue/static.rb:91:in 'CI::Queue::Static#poll'
  | from /tmp/bundle/ruby/3.4.0+0/gems/ci-queue-0.52.0/lib/minitest/queue.rb:229:in 'Minitest::Queue#run_from_queue'
  | from /tmp/bundle/ruby/3.4.0+0/gems/ci-queue-0.52.0/lib/minitest/queue.rb:214:in 'Minitest::Queue#__run

...
The test fail when ran alone, no need to bisect.

This PR gracefully handles the case when the failing test does not exist.

[casperisfine]

leak should always be triggered with the same version of the code where it happened, but I see no reason not to more gracefully handle this mistake.