Correct TEXCLUT addressing for CSM1 and CSM2

[GTTeancum]

Summary

• ignore TEXCLUT coordinates for CSM1's fixed CLUT layout
• apply TEXCLUT only to CSM2 lookups
• interpret CSM2 COU in its documented 16-pixel units
• cover CSM1 and CSM2 independently with regression tests

Rationale

The CPU sampler applied CBW, COU, and COV to both CLUT storage modes and treated COU as a pixel offset. CSM1 does not use TEXCLUT; CSM2 uses COU << 4 and COV to address its source rectangle.

Tests

• built ps2x_tests successfully with MSVC
• MINITEST_FILTER=PS2GS: both new tests pass
• the instrumented game-development checkout has one unrelated pre-existing CSR test failure; this branch is based directly on pristine upstream main