It is very much unsafe to do so as the string_ref could point on a stack
variable of the callee.

This is very much unsafe as the string_ref could point on a stack
variable of the callee.

- Had to chnage the core API to add a destroy function pointer in
  grpc_auth_metadata_processor.
- Tested end to end.
- Fixed some issues in the server_auth_filter (we were not checking the
  length which put us at risk of an overflow).

- Guarantee the lifetime of objects passed to the processor.
- Add an option for implementers of the processor to make it
  non-blocking.

- We always do the processing asynchronously but maintain a synchronous
  API for the implementor of the processor.
- there are a lot of string copies right now. Having a StringPiece
  object in grpc++ would really help with that (as we would use it for
  C++ metadata).
- Please review the API carefully and if you're happy with it, I'll
  proceed with tests.

to it.
	Change on 2015/01/09 by yangg <yangg@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=83640373