fix: superRefine() errors surfacing on optional() fields

[tushargr0ver]

Fixes #5653.

This PR updates ZodOptional to ensure validation chains (like superRefine) continue execution even if the value is undefined, correcting a regression where errors were swallowed.

[pullfrog]

Medium priority — This PR addresses a legitimate regression from #5589. The approach of distinguishing optout = "optional" (key can be absent) from optout = "includeUndefined" (value type includes undefined, but validation runs) is sound.

However, the PR lacks a regression test for the bug it claims to fix. A test like the one from issue #5653 should be included to prevent future regressions:

test("superRefine on optional field in object", () => {
  const schema = z.object({
    field: z.any().optional().superRefine((val, ctx) => {
      ctx.addIssue({ code: 'custom', message: 'should fail' });
    })
  });
  
  // Should fail - superRefine adds an issue even for absent keys
  expect(schema.safeParse({}).success).toEqual(false);
});

Additionally, util.optionalKeys still checks optout === "optional", which will now only match exactOptional() fields. This appears to be dead code (computed but never used), but it should either be updated for consistency or removed.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ pullfrog.com ｜ 𝕏}

[pullfrog]

This interface now extends $ZodTypeInternals directly instead of $ZodOptionalInternals. This is necessary because $ZodExactOptional needs optout = "optional" while $ZodOptional now has optout = "includeUndefined". The approach makes sense, but the duplication of fields (optin, optout, isst, values, pattern) could be avoided by having both interfaces extend a shared base.

Not blocking, but consider whether there's a cleaner way to express this inheritance relationship.