Add cfg to use __atomic_* builtins

[taiki-e]

• --cfg portable_atomic_unsafe_atomic_builtins
  
  Use __atomic_* builtins.

  Combine this with custom atomic logic and you can support CAS on multi-core systems where atomic CAS is not available in the standard library.

  Note: This cfg is unsafe.

  This cfg enables pointer-width and smaller atomic types.

  To enable atomic types grater than pointer-width, you need to enable --cfg portable_atomic_unsafe_atomic_builtins_N cfgs.

• --cfg portable_atomic_unsafe_atomic_builtins_N
  
  Similar to --cfg portable_atomic_unsafe_atomic_builtins, but also enables the specified size and smaller atomic types if the pointer width is smaller than the specified size.

  For example, when --cfg portable_atomic_unsafe_atomic_builtins_4 is enabled:

  • On 64-bit platform, Atomic{I,U}{64,32,16,8} will be enabled.
  • On 32-bit and 16-bit platform, Atomic{I,U}{32,16,8} will be enabled.

  N is in bytes and must be 4, 8, or 16. In other words, there are three valid patterns:

  --cfg portable_atomic_unsafe_atomic_builtins_4
  --cfg portable_atomic_unsafe_atomic_builtins_8
  --cfg portable_atomic_unsafe_atomic_builtins_16
  

cc #26, #33

[taiki-e]

Closing in favor of #51 for now.