Represent native empty, nullable, and operation-specific results accurately #392

Part of #382 — v1.0.0 alpha readiness.

Confirmed cases

The following signatures omit valid native states or attribute a property/result to the wrong object/operation:

• src/fetch/FetchTypes.res:11-L40 and src/core/DOM.res:82-L90: request destination/referrer-policy enums omit the empty string returned by default.
• src/media/DomHTMLMediaElement.res:3-L5: unsupported canPlayType returns an empty string that is absent from the closed result type.
• src/canvas/HTMLCanvasElement.res:117-L122: toBlob can invoke its callback with null, including a zero-sized canvas.
• src/device/WebLocksTypes.res:38-L45: an unavailable lock with ifAvailable: true supplies null to the callback.
• src/storage/IDBFactory.res:12-L18 and src/storage/IndexedDbTypes.res:125-L127: successful database deletion returns undefined, but the shared open-request result is fixed to a database.
• src/messaging/NotificationTypes.res:23: permission is static, not a Notification instance property.

The review's isolated Chromium probes confirmed these native outcomes; existing compilation alone does not validate them.

Acceptance criteria

• Add explicit empty-string variants where the native closed enum includes them, consistently across duplicate definitions.
• Make Canvas and Web Locks callbacks account for null without losing their successful-value types.
• Give IndexedDB open/delete requests operation-correct result types.
• Remove the false Notification instance field and retain/test the static accessor.
• Add runtime regressions for defaults, unsupported MIME types, zero-sized canvas encoding, contended optional locks, database deletion, and static permission access.
• Add compile/negative fixtures enforcing the corrected contracts and update affected examples.
• Preserve native failure behavior; do not solve these cases with unchecked casts or invented fallback values.

Coordination

Use the regression infrastructure in #383. Source fixes can be developed in parallel, but this issue should close with executed regression coverage, not only a successful compile.