Dir.glob brace pattern with '/' after '**' does not match
paths in recursive expansion process.
We expand braces with '/' before expanding a recursive.

Co-authored-by: Nobuyoshi Nakada <nobu@ruby-lang.org>

[Misc #18891]

Otherwise it's way too easy to confuse it with US_ASCII.

When using `IFTODT` defined in libc, `dirent.d_type` oriented pathtype
is compatible with `IFTODT(stat.st_mode)`. However they are not
compatible when emulating `IFTODT`, so `glob_helper` has to stat instead
of reusing dirent result by passing unknown pathtype to `glob_helper`.

This is a follow-up fix of 0c90ca4dd0abbd28d7bb34b9241d93995ab9cfb7

dir.c defines IFTODT if the system doesn't have it. The macro is used
when comparing with rb_pathtype_t's cases. rb_pathtype_t's cases are
defined by DT_XXX macro if they are available, or defined using IFTODT.
Most POSIX-compatible platforms have both IFTODT and DT_XXX and most of
other platforms like MinGW have neither of them. On those platforms,
DT_XXX-oriented rb_pathtype_t is always compared with values converted
by system's IFTODT, and emulated-IFTODT-oriented rb_pathtype_t is always
compared with values converted by emulated-IFTODT.

However, when IFTODT is *not defined* and DT_XXX is *defined*, like
on wasi-libc, DT_XXX-oriented rb_pathtype_t was compared with values
converted by emulated-IFTODT, and they are not guaranteed to be
compatible.

This patch fixes such a situation by using emulated-IFTODT to define
rb_pathtype_t when either IFTODT or DT_XXX is not available.