aco: don't optimize cross-lane instructions across p_wqm
We will use p_wqm as a marker in the next step.
Totals from 8846 (11.55% of 76572) affected shaders: (GFX11)
Instrs: 7031274 -> 7072729 (+0.59%); split: -0.02%, +0.60%
CodeSize:
37060272 ->
37355244 (+0.80%); split: -0.01%, +0.80%
VGPRs: 402660 -> 398724 (-0.98%); split: -0.99%, +0.01%
Latency:
62231926 ->
62322311 (+0.15%); split: -0.01%, +0.15%
InvThroughput:
10341361 ->
10392589 (+0.50%); split: -0.00%, +0.50%
VClause: 105344 -> 105368 (+0.02%); split: -0.03%, +0.05%
SClause: 218330 -> 218469 (+0.06%); split: -0.07%, +0.14%
Copies: 378609 -> 377644 (-0.25%); split: -0.42%, +0.17%
Branches: 97218 -> 97207 (-0.01%); split: -0.01%, +0.00%
PreSGPRs: 307654 -> 307644 (-0.00%); split: -0.08%, +0.08%
PreVGPRs: 314744 -> 308650 (-1.94%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25038>