Avoid extend/wrap to 64 bits when 32 are enough

[kripken]

E.g.

(i32.wrap_i64
 (i64.add
  (i64.extend_i32_u
   (local.get $0)
  )
  (i64.const 8)
 )
)
 =->
(i32.add
 (local.get $0)
 (i32.const 8)
)

This is an example of a more general rule: we could avoid extending and wrapping if the work in the middle will not matter after the wrapping.

Found by the superoptimizer #4994 (for comparison to other findings: rule #0, benefit 1560922).

[kripken]

Another example (rule #6):

(i32.wrap_i64
 (i64.mul
  (i64.extend_i32_u
   (local.get $0)
  )
  (i64.const 48)
 )
)
 =->
(i32.mul
 (local.get $0)
 (i32.const 48)
)