Remove yk_promote from MP_BC_LOAD_CONST_OBJ.

[Pavel-Durov]

Remove yk_promote from MP_BC_LOAD_CONST_OBJ

Calling yk_promote at MP_BC_LOAD_CONST_OBJ adds overhead on every bytecode constant load. This PR removes it.

perf diff

Symbol	Baseline (with yk_promote)	Delta (without yk_promote)
__yk_opt_gc_alloc	85.76%	-5.05%
mp_execute_bytecode	6.98%	+2.65%

Runtime accounting

  with-promote:    29.655s
  without-promote: 22.227s
  extra:            7.428s
  diff: 29.655 / 22.227 = 33% (with-promote is 33% slower)

Time in __yk_opt_gc_alloc

  with-promote:    29.655s * 86% = 25.5s
  without-promote: 22.227s * 81% = 18.0s
  extra:                           +7.5s

Time in mp_execute_bytecode

  with-promote:    29.655s * 6.8%  = 2s
  without-promote: 22.227s * 9.75% = 2s
  diff: 0

Conclusion:
__yk_opt_gc_alloc accounts for almost 100% of the runtime regression
mp_execute_bytecode shows a diff in perf (6.8% vs 9.75%) but the
absolute time is the same (~2s each)