Fix return type of encode_sample

[omus]

Broke off this non-breaking change from #163

[kleinschmidt]

is this non-breaking? it's possible currently working coudl could error if the results storage is allocated with a different type than S here.

I'm also not totally clear what the intended scope of this function is. it seems like it's only used internally, and the conversion is implicit in the normal use case of encode which pre-allocates results storage of the appropriate type if encoding is necessary...

[omus]

is this non-breaking? it's possible currently working coudl could error if the results storage is allocated with a different type than S here.

I'm also not totally clear what the intended scope of this function is. it seems like it's only used internally, and the conversion is implicit in the normal use case of encode which pre-allocates results storage of the appropriate type if encoding is necessary...

It's hard to know the author's intent on this non-exported, non-documented, non-tested function. I would say though based upon the first argument being a type the intent is to return a value of type S. Only in the case where S === Float32 could you see this issue come up and you are correct in practise those using encode! wouldn't see encounter any issue as an automatic convert happened for them. I do think this is non-breaking.

[omus]

@jrevels would be good to also get your review on this PR as you are the original author.