Use a thread for meminfo and defer tag lookup for copies

[unknownbrackets]

In some games, especially when using detailed meminfo, the main reason meminfo is "slow" is that it has to sometimes look up the current tag to tag the write. For example, a copy from "VideoDecode" via memcpy has the tag "ReplaceMemcpy/VideoDecode" which helps fix texcache issues AND is useful for debugging.

But to find the previous tag, it has to flush any pending tag data, and this is done for all sorts of copies. And in most cases, we don't immediately need the previous tag (except VideoDecode, etc. above.)

This puts the copy (where the tag isn't immediately needed) on the queue entirely, including the src tag lookup. This prevents the unnecessary flush, allowing it to flush later.

Additionally, this adds a thread that handles flushing (the meminfo tagging already needed to be threadsafe, anyway) so for write-only common cases, the main thread will not spend any time flushing. This does increase time spent in locks a little, but it still speeds up the main thread. Helps games that do a lot of medium size block transfers or memcpys.

-[Unknown]

[hrydgard]

Pretty decent boost at my favorite savestate in God of War: Chains of Olympus, 5%-ish or so, which is surprisingly much. It does do a lot of block transfers though.