fix: separate image extraction from schema parsing in content layer

[ascorbic]

Changes

Currently image imports in the content layer are extracted during schema parsing with the image() helper. This caused a problem when entries were reused from the data store, as they would need to be re-parsed if the images were to be found. This is what the glob loader was doing, causing #11737, where the transformed object does not match the schema.

This was quite a tricky one to fix, because of the way the image helper is so closely tied to zod parsing. The fix is to split the detection from the asset handling part: the image helper still prefixes the values with __ASTRO_IMAGE_ so that they can be identified in the runtime and replaced. However the image helper no longer registers the asset itself. Instead, the assets are registered when the entry is added to the store: the data is traversed, and any images are identified by their prefix, registered as assets and added as a new top-level field on the object. This means that re-used entries don't need to be re-parsed, as the assets are already known and can be re-registered instead.

Fixes #11737

Testing

Fixtures updated and new tests added.

Docs

[changeset-bot]

🦋 Changeset detected

Latest commit: aa43a77

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR