Add initial specs for IO::Buffer

[trinistr]

This class currently has no specs at all. This PR adds specs for the following methods:

• .new,
• #free, #resize, #transfer,
• all predicate methods,
• incidental tests for many other methods, including .map, .for, .strings and some instance methods.

Special remarks:

• spec for #valid? is somewhat incomplete, as garbage-collecting a string is non-trivial (I don't know how to);
• spec for #resize does not have tests for slices as currently it seems to be an undefined behavior, and it's not clear if it should be reproduced (yes, ruby/spec is all about reproducing even weird behavior, but bugs are a gray area still).

---

Weird things discovered so far so I don't forget:

• .new allows any and all flags (https://bugs.ruby-lang.org/issues/21672).
• Documentation is wrong (.map, for example, has very early text).
• INTERNAL and EXTERNAL are unclear, not being opposites (both can be false at the same time) and not even forming an exclusive tri-state with MAPPED.
• Slices feel undercooked:
  • No flags are copied from the source, except READONLY.
  • No way to detect a slice except to read #to_s (or call all predicate methods to check for all-false).
  • Is calling #free even safe on a slice? (io_buffer_free seems to work fine only because no flags are copied, not actually checking "is this a slice?").
  • Calling #free on the source buffer is not safe (https://bugs.ruby-lang.org/issues/21212).
  • And #lockeding does not propagate in either direction.
  • #freeing a String-backed buffer does not invalidate its slice (shouldn't slices always become invalid without a source buffer?).
  • Slices can be resized at any point?! (Breaks the slice if size is 0, otherwise creates a new internal buffer. (Again, this is a case of "not checking for slice, and no flags are set".))
• Is SHARED intended to be used manually or not?
• Why is PRIVATE mapping neither INTERNAL or EXTERNAL?
• .for and .string with a block are supposed to nullify buffer at the end, but buffer can be #transferred outside, is this intentional? (this creates a run-away buffer, able to change the string from wherever (but not a frozen string, at least)).
• #resize on a MAPPED buffer changes it to INTERNAL on macOS and Windows.

[trinistr]

This is very clearly a bug, so I filed it as such. Not sure if the test needs to be here. Probably not? This does not seem like behavior that should be replicated. I will delete it for now.