The previous timeout was short enough that it led to erroneous failures.

Fixes #1875.

That value actually MAY be NULL in some circumstances, and that's not a problem.

As a no-op for the time being.

- stops attempting to verify by catching the CancelledError

- instead the test examines the operation state after cancellation,
  which is equally valid and more stable.

Also
- simplifies bidi on the server

Introduced grpc_byte_buffer_init instead. It's now the user's responsibility to
manage memory.