test: use adapter-specific :on_duplicate_key_update shape

MySQL rejects the :conflict_target hash key (treats it as a column:
"Unknown column 'conflict_target'"). It infers the conflicting key on
its own, so it takes the bare column list, while PostgreSQL/SQLite need
the explicit conflict_target. Branch the option in the spec and document
both shapes in the README.

Author: Fivell

README.md

@@ -241,6 +241,7 @@ update colliding rows and insert new ones in a single pass with
 
 ```ruby
 ActiveAdmin.register Author do
+  # PostgreSQL / SQLite
   active_admin_import validate: false,
                       on_duplicate_key_update: {
                         conflict_target: [:id],
@@ -251,8 +252,16 @@ end
 
 Notes:
 
-* `conflict_target` names the unique column(s) used to detect a collision
-  (`[:id]` for the primary key). MySQL infers it and ignores this option.
+* The option shape is **adapter-specific**, since it is passed straight to
+  `activerecord-import`:
+  * PostgreSQL / SQLite need an explicit `:conflict_target` — the unique
+    column(s) used to detect a collision (`[:id]` for the primary key).
+  * MySQL infers the conflicting key, so pass just the column list and omit
+    `:conflict_target` (passing it raises `Unknown column 'conflict_target'`):
+
+    ```ruby
+    on_duplicate_key_update: %i[name last_name birthday]
+    ```
 * Turn `validate` off for id-based upserts. `activerecord-import` runs
   uniqueness validations against the very rows the upsert is about to overwrite,
   so a model-level `validates_uniqueness_of` would otherwise reject the update.

spec/import_spec.rb

@@ -228,14 +228,21 @@ def upload_file!(name, ext = 'csv')
         Author.delete_all
         Author.create!(id: 1, name: 'John', last_name: 'Doe', birthday: '1900-01-01')
 
+        # The option shape is adapter-specific: MySQL infers the conflicting key
+        # and only wants the column list, while PostgreSQL/SQLite need an explicit
+        # :conflict_target (see README).
+        on_duplicate_key_update =
+          if ActiveRecord::Base.connection.adapter_name.match?(/mysql/i)
+            %i[name last_name birthday]
+          else
+            { conflict_target: [:id], columns: %i[name last_name birthday] }
+          end
+
         add_author_resource(
           # Uniqueness validation runs against the rows the upsert is about to
           # overwrite, so it must be off for an id-based upsert (see README).
           validate: false,
-          on_duplicate_key_update: {
-            conflict_target: [:id],
-            columns: %i[name last_name birthday]
-          }
+          on_duplicate_key_update: on_duplicate_key_update
         )
         visit '/admin/authors/import'
         upload_file!(:authors_with_ids)