Add a real implementation of SRFI 124 based on Gambit's wills

[dpk]

The implementation behaves a bit weirdly — if you inspect the ephemeron-key in the REPL, the ephemeron won’t be broken until the datum also has no references anywhere else. I assume this is something to do with Gambit’s GC/REPL/object table/whatever, and I’m not about to start poking around in there. But in essence, it works:

> (define a (list 1 2 3))
> (define b (cons 4 5))
> (define e (make-ephemeron a b))
> (set! a #f)
> (##gc)
> (ephemeron-broken? e)
#t
> (ephemeron-key e)
#f
> (ephemeron-datum e)
#f

I’m also not quite certain this is a correct implementation of reference-barrier. Afaict even in a full implementation it’s just a void procedure meant to keep a certain value on the stack and therefore accessible?

(Also, like in Chibi, the ephemeron will never be broken if the datum refers to the key. This can’t be fixed without changes in Gambit, I think.)

[dpk]

Wait. Hold off on merging this for now — I think I’ve worked out how to get it to collect ephemerons with a cycle from datum to key, but I don’t have time to try it out right now.

[dpk]

Okay, that’s a bit of a hack, but it works. The weirdness with the REPL persists, but if you want to inspect an ephemeron’s contents, you can use an explicit call to write and it’ll break the ephemeron as usual later. Now mergeable if you can confirm that my intuition about reference-barrier is correct.

[vyzo]

LGTM modulo one minor comment.

I honestly don't know what to do about the reference-barrier either, void seems fine.

[vyzo]

thank you!