gl_rasterizer: Avoid scenario locking already owned mutex

gpu.TickWork() may lock the texture_cache and buffer_cache mutexes, which are owned by the thread prior to invoking TickWork().

Defer invoking gpu.TickWork() until the scope ends, where the owned mutexes are released.

dynarmic: Invalidate CPU cache on all cores

shader_recompiler: Include <bit> header when std::count{r,l}_zero is used

IC instructions now check for a need to halt execution

Follow-up fixes for NVFlinger rewrite

Revert "Memory GPU <-> CPU: reduce infighting in the texture cache by adding CPU Cached memory."

applets/swkbd: Add support for an updated inline software keyboard

Memory GPU <-> CPU: reduce infighting in the texture cache by adding CPU Cached memory.

First Nugget: Reworked Garbage Collection to be smarter [originally from Project YFC]

Rewrite of the NVFlinger implementation

shader_recompiler/dead_code_elimination: Add DeadBranchElimination pass

- Used by Naruto Ultimate Ninja Storm.