aee

Updated `base/message_loop/message_pump_io_starboard.cc` and `base/message_loop/message_pump_ui_starboard.cc` using https://github.com/youtube/cobalt/blob/c0dfb97eaf2742bd21f6054dae61a5328029046d/base/message_loop/message_pump_libevent.cc#L273 as a reference.

The failed checks are due to infra issues. ``` C:\BuildTools\VC\Tools\MSVC\14.34.31933\include\xtree(44): fatal error C1088: Cannot flush compiler intermediate file: 'C:\Users\ContainerAdministrator\AppData\Local\Temp2\_CL_3c56ab4bex': No space left on device ```

The `MessagePump::Delegate` interface changed. New: ``` Delegate::NextWorkInfo next_work_info = delegate->DoWork(); ``` Old: ``` bool did_work = delegate->DoWork(); TimeDelta next_delayed_work_time; did_work |= delegate->DoDelayedWork(&next_delayed_work_time); ``` The updated //base implementation simplifies the running...

Do I need to keep retrying the on_device checks that failed? Are they more flaky than the normal checks (which are also quite flaky)?