25ns overhead latency timers

Another round of win32 fixes and documentation.

Let's be really pedantic about the status of our submodules.

correct win32-related API usage

-) Fixed a few more (much more rare) race conditions on shutdown.
-) Fixed a degenerate case if we create a server but never start it.

Various Win32 fixes.

Fixing a very edge case where the tcp_client can crash due to race conditions on connection abortion.

-) tcp client and server should no longer starve waiting on orphans
-) proper server shutdown sequence to prevent use-after-free.

By the standard, std::function can only capture class functions, and not
class methods. For some reasons, this works under gcc however. Visual
Studio rejects it. std::mem_fn is proper template to use.

Fix parallel make + MINGW32 dependencies.

Add rdtsc profiling support, implement begin/end and tag hiding

This change fixes a parallel make which always failed with "cannot find
-lprotobuf".

For C++ static libraries, protobuf was already added as dependency, see
commit 5383062d.

Add google root cert to grpc/php docker image

fix: trace for grpc_call_start_batch only displayed lower 32bits of tag

Migrated Objective C library to C batch API

Fix flow control issue

Delete the old api

Fix for a crash seen in the wild

Conflicts:
	vsprojects/Grpc.mak

Stopgap fix for fling_stream_test timing out

remove unused function from grpc_csharp_ext

Ensure all unsecure tests are built on Windows

implement gpr_cpu_num_cores() and gpr_cpu_current_cpu() on Windows