Refactor channel_get_cell_queue_entry_size() to avoid an unreachable line for test coverage, and fix a nasty lurking memory bug in channel_flush_some_cells_from_outgoing_queue()

Fix return values from channel_flush_some_cells() to correctly count cells directly written by channel_flush_from_first_active_circuit()

Nick Mathewson authored 11 years ago and Andrea Shepard committed 10 years ago

Otherwise, when we authority try to do a self-test because of
init-keys, if that self-test can't be launched for whatever reason and
so we close the channel immediately, we crash.

Yes, this a silly way for initialization to work.

Implement scheduler mechanism to track lists of channels wanting cells or writes; doesn't actually drive the cell flow from it yet