Fixups

[kd-11]

• Remove an AMD workaround that is no longer needed with newer drivers (since 19.4.3)
• Fix texture cache search for flip source (should restore performance in some cases)
• Fix confirmed_range calculation for hw blit transfers (sw memory write barrier for flushing)
• Fix out-of-bounds transfers due to no bounds check when testing local resources for overlap.
• Fix clear failure due to null stencil mask triggering an assert.

Fixes #5869, #5854, #5824, #5898

[Parthorisian]

For [BLUS30443] Physics interaction speed has been slowed down, master_build8035 (recorded results for same place- top of Latria tower, one of piles with 5-6 vases - World-3-2) PR_build
P.S. Thanks for increased shader compilation speed(for [BLUS30443] and [BLUS30653])

[kd-11]

Pretty sure that is unrelated, unless DeS started failing to find flip image in the cache which is unlikely. You have to completely replace rpcs3 (including renaming the executable to rpcs3.exe) when testing for performance regressions sometimes. Could also be because this PR is not rebased on latest master.

[kd-11]

Actually looks like something is wrong with DeS, not sure when it started though. Number of RSX readbacks has gone up considerably.
EDIT: This problem has been present for some time but only affects RSX thread itself (a lot more self-intersection when reading index buffers). Non-trivial to fix (requires complex texture cache interaction).

[Parthorisian]

OK, thanks for explanation, ive cleaned all available caches and tested b8021, still awful performance of physic interactions. It seems speed of physic interactions was fixed by latest master(b8033)