1. The DX/DY coordinates for BitBLT are now signed again, fixes 8514/A cursor issues.
2. Made the busy/data available processor statuses more like the ATI one for more stability and no stalls, fixes Windows 3.x and IBM OS/2 1.x (possibly later ones too) that use the built-in 8514/A driver.
3. Made the mode switching clearer (through a note) regarding the shadow sets, this should make every program happy, I hope.
4. In the Mach32 series only, a first load of the CRT sets also initializes the GE offsets if the corresponding bit is set, fixes Mach8/32 mode tests in their eeprom and demoai with hdiload from ATI/IBM.