Significantly speed up nir_opt_copy_prop_vars AGAIN!

!20381 (merged) was my first attempt at this and while it provided some good gains it also cause some regressions as can be seen in #9326 (closed)

In this merge request I have revisited the problem an eliminated the original bottlenecks while also eliminating the bottlenecks caused by my previous attempt at speeding the code up. The problem CTS test now compiles faster than before we dropped the GLSL IR copy prop code and the Blender shader goes back to it's previous compile times before the last set of speedups for this code landed.

Overall for the series:

The Blender shader compile time in issue #9326 improves as folows:
251.29 seconds -> 9.90 seconds

The CTS test dEQP-GLES31.functional.ubo.random.all_per_block_buffers.20
improves as follows:

2.38 seconds -> 0.68 seconds