fix(ai): treat gpt-5.x-codex slugs as vision-capable

The document/lab-OCR vision gate excluded every `-codex` slug from the
GPT-5.x line as text-only. That is stale: the `codex/responses` wire reads
an `input_image` block on these slugs (docs/codex-protocol-spec.md plus the
current model lineup), and the 5.2-Codex line is vision-tuned. An operator
pinned to a `-codex` slug therefore resolved no vision-capable provider and
hit a 422 "No vision-capable AI provider is configured" when reading a
document with AI.

Drop the `-codex` exclusion from the gpt-5 branch so a codex-only chain
resolves a document-vision provider; keep the `-mini`/`-preview`
reasoning-slug guard. Lab-OCR on a `-codex` pin gains vision the same way.

src/lib/ai/__tests__/vision-capability.test.ts

@@ -71,7 +71,7 @@ describe("supportsVisionForConfig", () => {
     expect(supportsVisionForConfig("local", null)).toBe(true);
   });
 
-  it("reports vision for codex only on multimodal slugs", () => {
+  it("reports vision for codex on multimodal slugs, including -codex", () => {
     for (const model of [
       "gpt-5.5",
       "gpt-5.4",
@@ -79,6 +79,13 @@ describe("supportsVisionForConfig", () => {
       "gpt-4o",
       "gpt-4.1",
       "o1",
+      // The GPT-5.x `-codex` specialists are vision-capable: the wire reads
+      // an `input_image` block on these slugs (docs/codex-protocol-spec.md +
+      // OpenAI's current model lineup). The old "-codex is text-only"
+      // exclusion was stale.
+      "gpt-5-codex",
+      "gpt-5.3-codex",
+      "gpt-5.1-codex-mini",
     ]) {
       expect(supportsVisionForConfig("codex", model)).toBe(true);
     }
@@ -86,9 +93,6 @@ describe("supportsVisionForConfig", () => {
     expect(supportsVisionForConfig("codex", "gpt-4")).toBe(false);
     expect(supportsVisionForConfig("codex", "gpt-3.5-turbo")).toBe(false);
     expect(supportsVisionForConfig("codex", null)).toBe(false);
-    // The `-codex` specialist slugs are text-only despite the gpt-5 prefix.
-    expect(supportsVisionForConfig("codex", "gpt-5-codex")).toBe(false);
-    expect(supportsVisionForConfig("codex", "gpt-5.1-codex-mini")).toBe(false);
     // Text-only o-series variants are excluded for codex too.
     expect(supportsVisionForConfig("codex", "o1-mini")).toBe(false);
     expect(supportsVisionForConfig("codex", "o3-mini")).toBe(false);

src/lib/ai/vision-capability.ts

@@ -73,9 +73,10 @@ function openaiSupportsVision(model: string): boolean {
 /**
  * Codex (ChatGPT-OAuth) vision models. Image INPUT is reachable over the
  * `codex/responses` endpoint (docs/codex-protocol-spec.md §2b) on a multimodal
- * slug — no API key needed. The accepted GPT-5.x line is text+vision; the older
- * `-codex` specialist slugs and the legacy `gpt-4`-class non-multimodal pins
- * are text-only. The slug we test is the codex client's resolved working slug
+ * slug — no API key needed. The whole GPT-5.x line is text+vision, including
+ * the `-codex` specialist slugs (gpt-5.x-codex); only the legacy `gpt-4`-class
+ * non-multimodal pins are text-only. The slug we test is the codex client's
+ * resolved working slug
  * (cached or chain head), so a slug rotation onto a non-vision model degrades
  * to false rather than 400-ing the user at the image block.
  *
@@ -85,10 +86,14 @@ function openaiSupportsVision(model: string): boolean {
 function codexSupportsVision(model: string): boolean {
   const m = model.toLowerCase();
   return (
-    // GPT-5.x line is multimodal (text + vision) — but the `-codex` specialist
-    // slugs (e.g. gpt-5-codex) are text-only, so they must not claim vision or
-    // every lab-OCR upload burns a doomed image attempt before falling back.
-    (m.startsWith("gpt-5") && !m.includes("-codex")) ||
+    // GPT-5.x line is multimodal (text + vision) — INCLUDING the `-codex`
+    // specialist slugs (gpt-5.x-codex), which gained vision in the 5.2-Codex
+    // line (screenshots / diagrams / charts). Verified against
+    // docs/codex-protocol-spec.md (an `input_image` block on a gpt-5-codex
+    // request) plus OpenAI's current model lineup. The historical
+    // "-codex is text-only" exclusion was stale — the wire reads images on
+    // these slugs, so lab-OCR and document reads both succeed on a `-codex` pin.
+    m.startsWith("gpt-5") ||
     m.startsWith("gpt-4o") ||
     m.startsWith("gpt-4.1") ||
     m.startsWith("gpt-4-turbo") ||