aco: p_start_linear_vgpr doesn't always need exec mask

Foz-DB Navi21:
Totals from 1605 (1.21% of 132657) affected shaders:
CodeSize: 14023700 -> 14020320 (-0.02%)
Instrs: 2589881 -> 2589052 (-0.03%)
Latency: 22478420 -> 22473359 (-0.02%)
InvThroughput: 3851237 -> 3851092 (-0.00%)
Copies: 215316 -> 215438 (+0.06%); split: -0.39%, +0.44%

Allows more vcmpx usage.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23302>

src/amd/compiler/aco_ir.cpp

@@ -849,6 +849,7 @@ needs_exec_mask(const Instruction* instr)
       case aco_opcode::p_logical_end:
       case aco_opcode::p_startpgm:
       case aco_opcode::p_init_scratch: return instr->reads_exec();
+      case aco_opcode::p_start_linear_vgpr: return instr->operands.size();
       default: break;
       }
    }