Fill gaps in shared pattern contracts

- modal-workflows: full state machine, htmx pattern, validation rules
- go-api: REST conventions, URL naming, status codes, error format, list response
- import-export: streaming export 3-layer architecture with Go example
- CLAUDE.template.md: updated to include modals and REST API references

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

kit/patterns/import-export/contract.md

@@ -71,8 +71,53 @@ fmt.Sprintf("%.2f", price)
 - Escape embedded double-quotes by doubling them: `""`.
 - Use `encoding/csv` with `csv.Writer` and set `csv.Writer.Comma = ';'`; it handles quoting automatically.
 
+## Streaming Export Architecture (Go)
+
+For exports with potentially large row counts use a 3-layer streaming pattern.
+Never load all rows into memory before writing — stream directly to the response writer.
+
+```
+Handler → sets HTTP headers + writes BOM → calls Service
+Service → delegates to Repository with a row callback
+Repository → queries in batches → calls callback per row
+Handler/Service → csv.Writer.Flush() after all rows
+```
+
+```go
+// Handler
+func ExportCSV(c *gin.Context) {
+    c.Header("Content-Type", "text/csv; charset=utf-8")
+    c.Header("Content-Disposition", `attachment; filename="export.csv"`)
+    c.Writer.Write([]byte{0xEF, 0xBB, 0xBF}) // BOM
+
+    w := csv.NewWriter(c.Writer)
+    w.Comma = ';'
+    w.Write([]string{"ID", "Name", "Price"}) // header row
+
+    err := svc.StreamRows(ctx, filters, func(row Row) error {
+        return w.Write([]string{row.ID, row.Name, formatPrice(row.Price)})
+    })
+    w.Flush()
+    if err != nil {
+        // headers already sent — log only, cannot change status
+        slog.Error("csv export failed mid-stream", "err", err)
+    }
+}
+
+// Repository — batch fetch with callback
+func (r *Repo) StreamRows(ctx, filters, fn func(Row) error) error {
+    rows, err := r.db.QueryContext(ctx, query, args...)
+    // ... scan and call fn(row) for each row
+}
+```
+
+- Use `JOIN` in the repository query to avoid N+1 per row.
+- Batch size is optional; streaming row-by-row is fine for most datasets.
+- Always call `w.Flush()` after the loop — `csv.Writer` buffers internally.
+
 ## Error Handling
 
 - Import errors should map to clear user-facing messages.
-- Export errors after streaming starts must degrade gracefully (human-readable fallback).
+- Export errors after streaming starts must be logged server-side only — HTTP headers are already
+  sent and the status code cannot be changed mid-stream.