Fallback to use result of first branch when decoding :or and :orn

[ikitommi]

from #910:

I see. Here's one repro:

(m/decode [:int {:decode/math str}] 1 (mt/transformer {:name :math})) ; => "1"
(m/decode [:and [:int {:decode/math str}]] 1 (mt/transformer {:name :math})) ; => "1"
(m/decode [:or [:int {:decode/math str}]] 1 (mt/transformer {:name :math})) ; => 1

so, options being:

1. do nothing: it's a feature
2. use first branch in case none is valid after decode
3. use first branch where value is changed in decode

Not sure what is the most correct, 1 or 2 here 🤔

this PR implements option 2.

This is a BREAKING change as the behavior changes. But, I think the current implementation is wrong and this is (more) correct.

[opqdonut]

What about when the first transformer returns a nil? Could that happen?

I'm thinking of something like (m/decode [:maybe :int] false false-to-nil-transformer) ==> nil

Or perhaps even just (m/decode [:or nil :int] nil trans)?

[ikitommi]

I thought it wouldn't matter, as [:or nil :int] with nil value succeeds and doesn't use the fallback but here's a case where is fails:

(m/decode
 [:or
  [:string {:decode/remove-secrets (constantly nil)}]
  :keyword]
 "SECRET!"
 (mt/transformer {:name :remove-secrets}))
; => "SECRET!"

Fixed it so that value of the first branch is always used as fallback, regardless of value.

[opqdonut]

change looks good, I like the tests. one question though.