oficina / slides Public

fix: slide order is data too (codex 4 at slide level); export status agrees with the held bytes; docs

The deck's `slides` array is now the SET of its slides and their order
is an `order` key on each slide map — ruling 18 one level up, and the
same helper rather than a second idea (`ordered_maps` / `orderedMaps`,
`place_entry` / `placeEntry`). `move_slide` writes the key; `add_slide`,
`duplicate_slide` and `remove_slide` never rebuild the array. A slide
reordered while somebody types on it used to lose the typing for the
reason a moved block did: the slide map their text, notes and blocks all
hang off was deleted and replaced by a copy.

Tests in both implementations: two docs exchange updates, one reorders
while the other types into the slide being moved and into its notes;
both sides read the new order with the text intact. Plus the same story
over the wire in `internal/slidestest`. The view sorts by the key and
does not expose it, so no reader downstream changed — `every-block.yjs`
moved bytes and no `.view.json` moved. `changes.mjs` learned that a
slide's own `order` key is a strip re-sync rather than a stage refetch.

The export status page and the file now agree about the hour.
rastrillo/jobs forgets a finished job ten minutes after it ends and this
app holds the bytes for the hour it promised, so the page 404'd for
fifty minutes while `.pdf` at the same address still served. The page,
its polled fragment and the API's status document fall back to this
app's own record (`exportFallback`); `ForgetExportJobs` is the seam the
test uses to reach that state without waiting ten minutes.

AGENTS.md caught up with waves A and B: the write gate is `Handle.Check`
as well as `deck.Parse`, reading order has a seam of its own, ABI 2 and
both wasm artefacts are traps, offboarding is rechecked per request, and
the upload receipt has a table name. One false comment in
`fixtures/uploads.mjs` corrected — the app mints the staging key now, so
the client's hash is a claim and not an address.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>