plant: private Cargill Gouda tracker (/trackers/cargill)

Owner-only tracker for learning the Cargill Bioindustrial plant in Gouda:
schema plant (entries, links with confidence, pending refs), plant.apply_entries
bulk import, overview with clusters and flow diagram, kind tabs, glossary,
questions, detail pages and owner inline editing. Docs in docs/plant-tracker.md.

docs/plant-tracker.md

@@ -0,0 +1,81 @@
+# Plant tracker (`/trackers/cargill`)
+
+A private (owner-only) knowledge base about the Cargill Bioindustrial plant in Gouda:
+departments, products, process steps, equipment, glossary terms, questions, notes and roles
+(never colleague names). Logged-out visitors get a 404 on the page, its data and its actions
+(`ownerOnlyPrefixes` in `src/lib/server/owner.ts`).
+
+## Schema `plant` (migration `drizzle/20261026133106_plant`)
+
+- `plant.entries`: `kind`, `slug` (unique per kind), `title`, `summary`, `body` (markdown),
+  `parent_id` (any kind, e.g. a note under a note), `tags`, `sort_order`, `source`,
+  `confidence` (`confirmed` | `assumed` | `open`), `answered_at` (questions), timestamps.
+- `plant.links`: `from_id`, `to_id`, `relation` (`feeds_into`, `made_in`, `uses`, `part_of`,
+  `related`), optional `label`, `dashed`, `confidence` (default `confirmed`; `assumed` or
+  `open` links show dashed in the flow and with a "?" badge on entry pages).
+- `plant.pending`: parents and links whose target doesn't exist yet. Every `apply_entries` run
+  retries them, so a later batch that adds the target resolves them. Entry pages list them as
+  "pending".
+
+Kinds: `department`, `product`, `process_step`, `equipment`, `glossary`, `question`, `note`,
+`person_role`.
+
+How tags drive the UI:
+
+- `cluster` (departments): a section in the departments tab and a card in the overview, with
+  every unit below it (units with `parent: "cluster-ea"`, and their own units, like MP1's).
+- `inactive`: dimmed everywhere; the "hide inactive" switch hides them (off by default).
+- `overview` (notes): shown at the top of the overview.
+
+## Bulk import: `plant.apply_entries(jsonb)`
+
+Upserts by `(kind, slug)`, then sets parents and links by slug. Run it as the app user or the
+owner; for big files pipe SQL through stdin, e.g.
+`select plant.apply_entries($doc$ { ... } $doc$::jsonb);`
+
+The argument is either an object `{ "source": ..., "entries": [...], "links": [...],
+"delete": [...] }` or a bare array of entries.
+
+```json
+{
+	"source": "Walter White, batch 5 (2026-10-26)",
+	"entries": [
+		{
+			"kind": "department",
+			"slug": "414-kolom-1",
+			"title": "414 Kolom 1",
+			"summary": "Fat splitting column, continuous",
+			"body": "In and out:\n- In: feed, water and high-pressure steam.\n- Out: fatty acid and glycerine water.",
+			"parent": "cluster-dim",
+			"tags": ["splitting", "continuous"],
+			"confidence": "confirmed",
+			"links": [{ "to": "splitting", "relation": "uses", "label": "optional" }]
+		},
+		{ "kind": "question", "slug": "rndm", "title": "What is RNDM?", "tags": ["answered"] }
+	]
+}
+```
+
+- Required: `kind`, plus `slug` or a `title` to slugify; new entries need a `title`.
+- Optional: `summary`, `body`, `parent` (alias `parent_slug`; `null` clears it), `tags`,
+  `confidence` (default `assumed`, `open` for questions), `sort_order`, `source`, `answered`,
+  `replace_links`, `replace_tags`, `body_mode` (`"append"`), `links` (`to`, `relation`,
+  `label`, `dashed`, `confidence`, `remove`).
+- Updating an existing entry:
+  - Keys that are left out keep their current value. An empty `summary` or `body` also keeps
+    the current text. With `"body_mode": "append"` the new body is added below the current
+    one (skipped when it is already there) instead of replacing it.
+  - `tags` are merged into the existing tags, unless the entry sets
+    `"confidence": "confirmed"` or `"replace_tags": true`, which replace them.
+  - The top-level `source` is only the default for new entries; pass `source` on an entry to
+    overwrite it.
+- A question tagged `answered` (or with `"answered": true`) is marked answered; the tag is
+  dropped.
+- References (`parent`, `to`, `from`) are a slug or `kind:slug`. A bare slug used by several
+  kinds resolves in this order: process_step, product, department, equipment, note,
+  question, person_role, glossary (so `splitting` is the process step).
+- Unknown references never abort the import: they are listed in `unresolved` and kept in
+  `plant.pending` until their target exists.
+
+Returns `{ inserted, updated, linked, unlinked, deleted, unresolved: [{from, field, to}],
+resolved_pending, pending, entries: [{id, kind, slug, action}] }`.