Why does type inference for lambda differ from standard declaration?

In GHCI, if I use :t on each of the below functions, they differ.

:t test gives (Ord p, Num p) => p -> p -> p

:t testAsLambda gives Integer -> Integer -> Integer

Why does using a lambda to define the exact same function cause more restrictive type inference?

I understand that it might be silly to declare a named function as a lambda like this; I'm just trying to reason about Haskell's type inference here.

test x y =
  let d = y - x in
   if d > 0 then d
   else 0

testAsLambda = 
  (\x y -> 
   let d = y - x in
    if d > 0 then d
    else 0)