feat(labs): add local-OCR text mode for image-blind AI providers

Add an opt-in in-browser OCR path so a user whose provider cannot read
images (ChatGPT-OAuth/Codex, a text-only model) can still scan a lab
report. The capability probe gains a `mode` field: `vision` when the
provider reads the image directly, `text` when local OCR is enabled and
any provider is configured, or null. The extract route grows a text
branch that accepts the in-browser-OCR'd text as JSON and structures it
through the existing schema + review/commit flow; the raw image never
reaches the server in text mode. New `labsLocalOcrEnabled` user column
(migration 0184, default off) gates it, with a dedicated PATCH endpoint.

prisma/migrations/0184_v11810_labs_local_ocr/migration.sql

@@ -0,0 +1,16 @@
+-- v1.18.10 — opt-in local (in-browser) OCR for lab-report scans.
+--
+-- One additive, defaulted column on `users`. When `true`, a user whose AI
+-- provider cannot read images (ChatGPT-OAuth/Codex, a text-only model) can
+-- still scan a paper report: the image is OCR'd in the browser via tesseract.js
+-- and only the extracted TEXT is forwarded to the text-only provider for
+-- structuring. The raw image never leaves the device in this mode.
+--
+-- Less accurate than a native vision model — the mandatory per-row review
+-- screen stays the safety backstop, and the in-app copy sets that expectation.
+-- The default `false` backfills every existing row onto the current behaviour
+-- (vision-only scan, or no scan affordance), so this is a no-op for anyone who
+-- never opts in.
+
+ALTER TABLE "users"
+  ADD COLUMN "labs_local_ocr_enabled" BOOLEAN NOT NULL DEFAULT false;

prisma/schema.prisma

@@ -207,6 +207,14 @@ model User {
   // diagnosis — purely a user-declared preference that selects which target
   // band the user is judged against. Default `false`.
   hasDiabetes             Boolean @default(false) @map("has_diabetes")
+  // v1.18.10 — opt-in local (in-browser) OCR for lab-report scans. When `true`,
+  // a user whose AI provider cannot read images (e.g. ChatGPT-OAuth/Codex, a
+  // text-only model) can still scan a report: the image is OCR'd in the browser
+  // via tesseract.js and only the extracted TEXT is sent to the text-only
+  // provider for structuring. Less accurate than native vision — the mandatory
+  // review screen remains the safety backstop. Default `false`; vision-capable
+  // users keep the higher-accuracy native path and never pay the WASM download.
+  labsLocalOcrEnabled     Boolean @default(false) @map("labs_local_ocr_enabled")
   // v1.7.0 unit pref — global metric/imperial display preference.
   // Canonical storage stays SI; this only selects the display-time
   // transform branch (km/h vs mph, km vs mi). Null = default "metric".