This change poisons the whole slot of the object rather than just the flags.
This allows ASAN to find any reads/writes into the slot after it has been
freed.

Newer GCCs have __has_feature and older ones have
__SANITIZE_ADDRESS__[1]. Relevant since ASAN with GCC 11 on the popular
Ubuntu Jammy failed to build previously.

[1]: https://gcc.gnu.org/onlinedocs/gcc-4.8.0/cpp/Common-Predefined-Macros.html

There's no case for when RUBY_MSAN_ENABLED, so the macro ends up doing
nothing when it should instead have __attribute__((__no_sanitize__("memory"))).

```
internal/sanitizers.h:57:26: error: ISO C does not allow extra ‘;’ outside of a function [-Wpedantic]
   57 |     COMPILER_WARNING_PUSH; \
      |                          ^
```

and so many.

Remove semicolons following pragma, and repeat the given declaration
at the end to consume a semicolon following the macro call.  As many
`NO_SANITIZE` calls including bigdecimal that is a gem have a trailing
semicolon, it was not able to move the semicolon inside `NO_SANITIZE`.

Otherwise trying to deref the pointer can cause an ASAN crash, even
though the only reason we're dereferencing it is so that we can assert
on it.

ASAN leaves a pointer to the fake frame on the stack; we can use the
__asan_addr_is_in_fake_stack API to work out the extent of the fake
stack and thus mark any VALUEs contained therein.

[Bug #20001]