I want to apply a transformation to everything I send up my websocket, so I thought I could pass a channel with a mapping transducer as write-ch.

This doesn't work as I'd hoped because my transducer sees messages after formatting, not before. So I have to thaw, apply the transformation, and freeze again in my transducer.

Ideally, I think the formatting would occur as the last thing before writing to and the first thing after reading from the websocket connection - so read-from-ws! and write-from-ws! would take a formatter, and apply thaw and freeze respectively, rather than involving core.async at all.

That way the user-supplied channels will always be dealing with unformatted values, which I suspect will be what people want/expect almost all the time, and we lose a bit of core.async overhead inside of chord. This would break backwards compatibility in that the change could break anyone passing a channel with a transducer as write-ch. However I suspect most uses of read-ch and write-ch are to apply buffering as all mine were until today :)

Happy to make this pull request if you like the idea!