Fix precedence of stack_ keyword

[dvulakh]

#85 added support for the stack_ keyword. Internal adoption has revealed some bugs in the parenthesization behavior. Two examples:

Foo (stack_ ((), ()))

becomes

Foo stack_ ((), ())

(invalid syntax)

stack_ (() :: [])

becomes

stack_ () :: []

(parsetree change)

This PR fixes the precedence associated with Pexp_stack in the extended AST for these two patterns, as well as for a number of patterns that should be uncommon (because there stack_ does not enclose an allocation, and so the code is rejected by the typechecker). We should still aim to support these cases, because:

1. We should succeed in formatting any code that parses. If someone writes one of these forbidden patterns, they should get a (ostensibly helpful) error from the typechecker, not a mysterious formatting failure.
2. These patterns could be used in ppx payloads and then transformed to code that actually compiles.

It might be worth testing ocamlformat's treatment of stack_ more thoroughly in the near future, but I think it makes sense to merge this PR before doing so, since it fixes problems people have actually started to encounter.