refactor(explore): simplify form_data parse + add no-datasource regression

/review-code on the post-round-5 slice flagged two minor issues genuinely
introduced by 89919a1c96:

* `loads_request_json` already swallows `TypeError`/`JSONDecodeError` and
  returns `{}`, so the explicit `try/except ValueError` wrapping it was
  dead defensive code. Replace it with an `isinstance(..., dict)` guard
  — that also closes the pre-existing `?form_data=42` edge case where a
  non-dict top-level JSON value would `AttributeError` on `.get()`.

* The no-datasource fall-through path (the loop the fix unblocked) had
  no targeted regression test. `test_slices` and
  `test_slice_id_is_always_logged_correctly_on_web_request` exercised
  it only indirectly via `Slice.slice_url`. Add a sibling to
  `test_explore_redirect` that asserts `/explore/?form_data={"slice_id":1}`
  returns 200 (SPA render), not 302.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tests/integration_tests/core_tests.py

@@ -882,6 +882,16 @@ class TestCore(SupersetTestCase):
         rv = self.client.get(f"/explore/?form_data={quote(json.dumps(form_data))}")
         assert rv.headers["Location"] == f"/explore/?form_data_key={random_key}"
 
+    def test_explore_no_datasource_renders_spa(self):
+        # `Slice.slice_url` emits form_data carrying only `slice_id`; without a
+        # datasource the cache-and-redirect contract can't produce a different
+        # URL, so ExploreView.root must fall through to the SPA instead of
+        # 302-looping back to itself.
+        self.login(ADMIN_USERNAME)
+        form_data = {"slice_id": 1}
+        rv = self.client.get(f"/explore/?form_data={quote(json.dumps(form_data))}")
+        assert rv.status_code == 200
+
     @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices")
     def test_has_table(self):
         if backend() in ("sqlite", "mysql"):