chore(ci): replace toys kokoro-ci with a Ruby script

[torreypayne]

Part of the move off toys (b/563024235).

• Adds bin/kokoro-ci, a plain-Ruby port of .toys/kokoro-ci.rb that needs only default gems, and runs it from .kokoro/system_tests.sh.
• The presubmit rule that runs every product when test drivers change now also matches bin/.
• Deletes the unused update-gemfile-locks tool, which empties .toys/.
• Fixes three bugs copied from toys: cleanup after a failed setup, an early Ruby exit, and the SQL Server proxy check.

Parity: a harness ran both tools through 38 scenarios on Ruby 3.2 to 4.0 and recorded identical commands, environments and exit codes, apart from those fixes.

[gemini-code-assist]

Code Review

This pull request replaces the toys kokoro-ci tool with a pure Ruby script bin/kokoro-ci that runs using only default gems, updating .kokoro/system_tests.sh and removing the old toys tasks. The review feedback highlights three key issues in the new script: a potential resource leak during cleanup if gimmeproj done fails, an unhandled Errno::EPIPE exception when writing to a crashed Ruby process pipe, and a copy-paste bug that checks for PostgreSQL instead of SQL Server when starting the SQL Server proxy.

[gemini-code-assist]

If gimmeproj done fails during cleanup, the script will exit immediately because @exit_on_failure is still true (if setup failed). This prevents the subsequent background processes in @kill_on_cleanup from being killed, leading to leaked resources on the CI runner. Setting @exit_on_failure = false at the start of run_cleanup ensures that all cleanup steps are executed.

  # Ensure background processes are always killed even if gimmeproj fails.
  def run_cleanup
    @exit_on_failure = false
    run ["gimmeproj", "-project", "cloud-samples-ruby-test-kokoro", "done", @project] unless @project.empty?
    @kill_on_cleanup.each do |pid|
      Process.kill "SIGKILL", pid if pid
    rescue StandardError
      nil
    end
  end

[gemini-code-assist]

If the spawned Ruby process exits or crashes before reading the script from the pipe, writer.write will raise Errno::EPIPE (broken pipe). Since this exception is unhandled, it will crash the entire CI run instead of marking the individual test as failed. Wrapping the write and close operations in a begin ... rescue Errno::EPIPE ... ensure block prevents this crash and allows the script to correctly capture the process exit status.

  def run_ruby args, script
    reader, writer = IO.pipe
    pid = start_process [RbConfig.ruby, *args], argv0: "ruby", log: "exec ruby: #{args.inspect}", in: reader
    reader.close
    begin
      writer.write script if pid
    rescue Errno::EPIPE
      # Child process exited before reading the script
    ensure
      writer.close rescue nil
    end
    check_result pid && Process.wait2(pid).last
  end

[gemini-code-assist]

This is a copy-paste bug carried over from the original toys tool. It checks for cloud-sql/postgres to start the SQL Server proxy, which means the SQL Server proxy is started unnecessarily when running Postgres tests, and would not be started if only SQL Server tests were run. We should check for cloud-sql/sqlserver instead.

    if @products.include? "cloud-sql/sqlserver"
      say "Starting Cloud SQL Proxy for SQL Server", :bold
      connection_name = assert_env "SQLSERVER_INSTANCE_CONNECTION_NAME"
      pid = run_background ["/bin/cloud_sql_proxy",
                            "-instances=#{connection_name}=tcp:1433",
                            "-credential_file=#{gac_path}"]
      ENV["SQLSERVER_CLOUD_SQL_PROXY_PROCESS_ID"] = pid.to_s
      @kill_on_cleanup << pid
    end