lima/ppir: increase usage of ppir_pipeline_reg_fmul/ppir_pipeline_reg_vmul

Small optimization which should save some pp instructions/registers by swapping arguments of sadd/vadd block if possible.

This could be done in other ways too (ppir_instr_insert_mul_node?), so let me know what you think.

No regressions were observed with piglit. No dEQP run was done yet.