Unity 4.2 staging

[QNX] socket accept gets interrupted by real-time signals.  Loop over ac...

Unity 4.2 staging

Work around this for now by manually enabling -Os optimizations from the llvm doc.

Do not set thread priority to realtime as it causes hangups on iOS singl...

Do not set thread priority to realtime as it causes hangups on iOS single core devices. Proper fix pending.

it breaks re-attaching, needs a better fix (possible race condition here)
Fixes #539110

This reverts commit 88a3c6ea.

[QNX] socket accept gets interrupted by real-time signals.  Loop over accept function if failure cause was a signal interrupt

Fix libMonoPosixHelper.dylib for the editor

Android (general) fixes, BB10, and GC threads

This reverts commit 2feb13aa.

Enable thumb support for BB10, improve NDK auto download to use internal...

Don't push threads which are not suspended on Win32. Fixes a race condition described on GC mailing list where a new thread is being created during suspension process. Should avoid another cause of GetThreadContext failues.

shared handles; don't install atexit(handle_cleanup) if shared handles are disabled (atexit() is called at process termination; not at dlclose() - which fails on platforms with a dynamically loaded mono lib)

bsearch behavior is undefined when 'key' param is a null-ptr (on android 'undefined' sometimes means calling abort)

Unity staging

ARM: Fix issue when passing many and big value types (structs) from native code to managed code as method arguments, would cause memory overwrite and crash later. The change more properly estimates space required to store arguments in DynCallArgs.regs[] array. (Case 502916)

Add GC_stop_world_external and GC_start_world_external that properly disable signals and acquire GC locks before calling GC_stop_world and GC_start_world respectively. Use these new functions external of GC in liveness code. May fix case 530367.

	The timeout was not being used in GetResponse.
	Fixes bug #603794.

	In some weird cases socket is null when we try to close it.

	Send a Set-Cookie header per cookie instead of a Set-Cookie2 with
	multiple cookies.

	Fixes bug #667550.