This PR optimizes defined instruction for the DEFINED_YIELD case, which is used by defined?(yield).

It has been raised as an idea to use defined?(yield) instead of block_given? when we rewrite a C method in Ruby so that it will not be impacted by method redefinition.

Matz didn't ask for tolerating such redefinition, so we don't need to do it for compatibility. However, it could be still useful for performance when we rewrite C methods in Ruby since defined?(yield) is not a method call unlike block_given?, so it's faster on the interpreter.

Even for YJIT, defined instruction is more preferable because you don't need to split blocks for defer_compilation (smaller memory overhead) or add a guard for the receiver (faster).

Since we're interested in removing C -> Ruby calls, we would naturally rewrite methods that check block_given?. To make it easier for us to merge such changes, being able to use this optimization for the interpreter performance seems useful.

Example code

block_given?

  # Insn: 0001 opt_send_without_block (stack_size: 1)
  # call to Object#block_given?
  # guard known object with singleton class
  0x559640a680bf: movabs rax, 0x7f9f7f6ace00
  0x559640a680c9: cmp rsi, rax
  0x559640a680cc: jne 0x559640a6a0aa
  # block_given?
  0x559640a680d2: mov rax, qword ptr [r13 + 0x20]
  0x559640a680d6: mov rax, qword ptr [rax - 8]
  0x559640a680da: test rax, rax
  0x559640a680dd: mov eax, 0x14
  0x559640a680e2: mov ecx, 0
  0x559640a680e7: cmove rax, rcx
  0x559640a680eb: mov rsi, rax

defined?(yield)

  # Insn: 0001 defined (stack_size: 1)
  # block_given?
  0x559640a6805d: mov rax, qword ptr [r13 + 0x20]
  0x559640a68061: mov rax, qword ptr [rax - 8]
  0x559640a68065: test rax, rax
  0x559640a68068: movabs rax, 0x7f9f7f671940
  0x559640a68072: mov rax, rax
  0x559640a68075: mov ecx, 4
  0x559640a6807a: cmove rax, rcx
  0x559640a6807e: mov rsi, rax