Use add_command_buffer_generation_task for transparent and transmissive 3D passes

[fairhill1]

Objective

• Fixes Consider using add_command_buffer_generation_task() for MainTransparentPass3dNode and MainTransmissivePass3dNode #18305
• Improve CPU parallelization during render command generation for transparent and transmissive 3D passes
• Move command buffer generation work to separate tasks/threads instead of blocking the main render thread

Solution

Refactors MainTransparentPass3dNode and MainTransmissivePass3dNode to use the async command buffer generation pattern with add_command_buffer_generation_task(), matching the approach already used by MainOpaquePass3dNode.

Changes:

• MainTransparentPass3dNode: Convert to async pattern with command encoder
• MainTransmissivePass3dNode: Convert to async pattern while preserving screen space specular transmission steps logic
• Both nodes now follow the same performance optimization pattern used by the opaque pass

This change should improve performance by allowing render command generation to happen in parallel rather than sequentially on the render thread.

Testing

• Needs benchmarking
• Tested with existing 3D rendering examples to ensure no visual regressions
• Verified that transparent and transmissive materials continue to render correctly
• Confirmed that the async command buffer generation works as expected
• No compilation errors or runtime issues observed

Reviewers can test by:

1. Running any 3D examples with transparent materials (e.g., transparency_3d)
2. Running examples with transmissive materials to verify screen space transmission still works
3. Checking that performance characteristics match the opaque pass behavior
4. Verifying no visual differences in rendered output

[JMS55]

A little hard to review the diff, but I think this looks good.

[IceSentry]

This should be an early return to reduce nesting. Nothing else happens if this check is false

[IceSentry]

The descriptor should be constructed once before the if steps > 0. It's the same in both branches. That would probably help reduce the diff size too.

[IceSentry]

A few nitpicks for the transmissive pass that should help make the code shorter/easier to review