Add the option of adding a non-listening server completion queue. This makes writing certain test cases (like hybrid_end2end tests) easier

This reverts commit 5b304753.

Initial interface rework to allow knowing whether to pull payload at registration, not at request time

Allows the bottom member of the stack to schedule the actual deallocation, allowing a final transport lock to be entered when destroying a call.

- Have the server clear the accept_stream callback prior to destroying
  the channel (required a small transport op protocol change)
- Have the transport not enact transport ops until parsing is completed
  (prevents accept_stream from disappearing mid-parse)

Allow plugins to extend the set of filters used by gRPC core:
- plugins at construction time can register against the 'channel_init' system to be allowed to mutate a new channel_stack_builder type
- channel_stack_builder provides a central and rather dynamic place to construct the list of filters required by a channel stack
- ultimately we construct the channel stack in the fashion we always have

This is also a prerequisite step to allowing filters to be implemented from wrapped languages.

Specifically:

Receiving trailing and initial metadata had to be published in
lock-step.
=> If we wanted trailing metadata, we might not get initial metadata processed
   until messages arrived.
=> Compression code had no idea what codec to use.

To fix it, publish initial metadata as soon as it's ready (this is a
transport API change).

Requires changes to grpc_call to ensure ordering in processing initial
metadata and messages (one may be delayed).

Exposed at least some bugs in C++ where we never read initial metadata.

I expect at least one more similar bug.

- cleanup: change grpc_iomgr_cb_func to take a bool instead of int
  success
- cleanup: follow through with iomgr callback scheduling functions
- prepare: add a workqueue to offload to to grpc_exec_ctx_enqueue*
  functions

Fix the initialization by modifying the grpc_channel_args_is_census_enabled to by default return census_enabled() instead of always disable

Fix the census initialization on the server. Remove duplicate initialization logic and also enable census per channel if census is initialized.

Some transport callbacks weren't getting triggered. We now ensure that
this always occurs.

Additionally: left some debug only code around to make this class of bug
easier to diagnose in the future.