feat(api): OCR réel — POST /ocr/jobs + GET /ocr/jobs/:id (tesseract syscall, jobs asynchrones)

- ocr/tesseract.go : Engine → Tesseract subprocess (OCR_LANG fra+eng) ; PDF → calque texte via ledongthuc/pdf (go.mod : nouveau dep)
- repository/ocr_jobs.go : queued→processing→done/failed, scoping device, text/error NULLIFés
- service/ocr.go : Create valide le fichier (GetFile), queue + goroutine de traitement ; physique résolu par glob UPLOAD_DIR/<device>/<id>.* ; fail propre (fichier illisible, erreur moteur)
- handlers/ocr.go : réels (validate fileId, NOT_FOUND si job d'un autre device), fin 501 OCR
- tests end-to-end : cycle queued→done (stub moteur), NOT_FOUND fichier inconnu, scoping device
- smoke réel : PNG 'VAULTDROP' → job done text='VAULTDROP' (tesseract installé)
- docs/AGENTS : §5 + état des routes (tout V1 réel)

backend/ocr/tesseract.go

@@ -0,0 +1,67 @@
+package ocr
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"os/exec"
+	"path/filepath"
+	"strings"
+
+	"github.com/ledongthuc/pdf"
+)
+
+// TesseractEngine runs the `tesseract` binary in a subprocess.
+// Images are OCR'd directly; PDFs have their text layer extracted first
+// (scanned PDFs → empty text, no rendering pipeline in V1).
+type TesseractEngine struct{}
+
+func NewTesseract() *TesseractEngine { return &TesseractEngine{} }
+
+// ExtractText implements Engine.
+func (t *TesseractEngine) ExtractText(ctx context.Context, filePath, lang string) (string, error) {
+	if strings.ToLower(filepath.Ext(filePath)) == ".pdf" {
+		return extractPDFText(filePath)
+	}
+	return runTesseract(ctx, filePath, lang)
+}
+
+func runTesseract(ctx context.Context, filePath, lang string) (string, error) {
+	cmd := exec.CommandContext(ctx, "tesseract", filePath, "stdout", "-l", lang)
+	var stdout, stderr bytes.Buffer
+	cmd.Stdout = &stdout
+	cmd.Stderr = &stderr
+	if err := cmd.Run(); err != nil {
+		if errors.Is(ctx.Err(), context.Canceled) {
+			return "", ctx.Err()
+		}
+		msg := strings.TrimSpace(stderr.String())
+		if msg == "" {
+			msg = err.Error()
+		}
+		return "", errors.New("tesseract: " + msg)
+	}
+	return strings.TrimSpace(stdout.String()), nil
+}
+
+func extractPDFText(filePath string) (string, error) {
+	f, r, err := pdf.Open(filePath)
+	if err != nil {
+		return "", errors.New("pdf: " + err.Error())
+	}
+	defer f.Close()
+	var builder strings.Builder
+	for i := 1; i <= r.NumPage(); i++ {
+		p := r.Page(i)
+		if p.V.IsNull() {
+			continue
+		}
+		plain, err := p.GetPlainText(nil)
+		if err != nil {
+			continue
+		}
+		builder.WriteString(plain)
+		builder.WriteString("\n")
+	}
+	return strings.TrimSpace(builder.String()), nil
+}