feat(plugin-import-export): give import/export hooks access to the triggering document

[nathanlentz]

What

Gives the collection-level import/export hooks access to the document that triggered the run, as exportDoc and importDoc. This lets a hook read a value an editor set on the import/export form — including a field the project added with overrideCollection — and apply it to every row.

importExportPlugin({
  collections: [
    {
      slug: 'posts',
      export: {
        overrideCollection: ({ collection }) => {
          collection.fields = [...collection.fields, { name: 'batchRef', type: 'text' }]
          return collection
        },
        hooks: {
          before: ({ data, exportDoc }) =>
            data.map((row) => ({ ...row, batchRef: exportDoc.batchRef })),
        },
      },
    },
  ],
})

How

The document is read once when the run starts and passed unchanged to every batch, so hooks see a snapshot from when the operation was triggered rather than live state. A large export can fire the hook hundreds of times, so re-reading per batch would be both wasteful and inconsistent mid-run.

Where the document comes from depends on the path:

Path	Source	Has id
Export via jobs queue	findByID in the task handler	yes
Import, synchronous and jobs queue	the saved document both callers already hold	yes
Export, synchronous	submitted form data, in beforeOperation	no
Export download	submitted request body	no
Export and import preview	submitted request body	no

id is therefore optional, and an id means the document is saved. To keep that rule true rather than merely conventional, the paths without a saved document run the incoming body through getSubmittedFormValues, which drops id, createdAt and updatedAt. A caller cannot put an id on a download or preview request and have a hook treat the run as saved.

The two preview components previously built their request bodies from a fixed list of built-in fields, so a custom field never reached the endpoint. Both now forward the whole form. A value-based signature of the form state is added to the debounce dependencies, so editing a custom field refreshes the preview once the editor stops typing rather than on every keystroke.

Tests

• Integration coverage for before and after on every path: synchronous and queued import, queued and synchronous export, and download.
• Document identity across batches, for both import and export.
• Two tests where the form value actually changes what gets written — a stamped column on export, a modified title on import.
• Two tests that a submitted id cannot make an unsaved run look saved.
• Two Playwright tests that drive the real admin forms. These are the only tests that can catch a hand-built request body, since an integration test builds its own.

A posts-with-hooks-jobs collection was added so both the synchronous and task-handler paths are exercised; its plugin config leaves the jobs queue enabled.

Open item

Import and export previews also render on a saved operation document's edit view, where useDocumentInfo() has an id. The clients do not send it, so hooks previewing a saved document still receive an unsaved snapshot. Fixing it means sending the route ID and fetching the document server-side, and needs a decision on what happens to the open form's unsaved edits — merging them keeps the preview live but means an id no longer implies every value was validated. Left out of this PR deliberately.

[github-actions]

📦 esbuild Bundle Analysis for payload

This analysis was generated by esbuild-bundle-analyzer. 🤖
This PR introduced no changes to the esbuild bundle! 🙌