i.MX51 GPU3D command processing and A2xx rendering

[cavenderbi]

Implement the i.MX51 GPU3D command and software rendering paths needed by supported Ford SYNC 2 Navigation workloads. Guest commands now run through checked memory and packet handling, execute A2xx shaders, and rasterize supported geometry into logical GMEM for resolve to guest memory.

• Bound ring and indirect-buffer processing, validate physical-memory spans, and support context shadowing and blits.
• Connect indexed draws to clipping, triangle rasterization, depth16 testing, blending, tiled resources and supported non-face BIN visibility.
• Execute supported A2xx arithmetic, predicates, conditional clauses, nested calls and loops, relative operands and MOVA address conversions. Discard buffered fragment exports when the invocation is killed.
• Add quad-based shader derivatives, explicit texture gradients, supported mip/cube sampling and LOD/filter/border queries. Apply scalar predicate priority and legacy zero-input multiplication behavior.
• Save and restore modeled GPU context, draw and GMEM state.

The GPU2D constant-blend and keyboard color-transform changes are separate in #26.