FIX: gracfully handle missing seaborn dependency (pydata#9835)

xref: pydata#9561 (comment)

Fixes regression introduced in pydata#9561

this specific seaborn import should be handled gracefully, as it was before #pydata#9561

xarray/plot/utils.py

@@ -143,14 +143,11 @@ def _color_palette(cmap, n_colors):
             pal = cmap(colors_i)
         except ValueError:
             # ValueError happens when mpl doesn't like a colormap, try seaborn
-            if TYPE_CHECKING:
-                import seaborn as sns
-            else:
-                sns = attempt_import("seaborn")
-
             try:
-                pal = sns.color_palette(cmap, n_colors=n_colors)
-            except ValueError:
+                from seaborn import color_palette
+
+                pal = color_palette(cmap, n_colors=n_colors)
+            except (ValueError, ImportError):
                 # or maybe we just got a single color as a string
                 cmap = ListedColormap([cmap] * n_colors)
                 pal = cmap(colors_i)
@@ -192,7 +189,10 @@ def _determine_cmap_params(
     cmap_params : dict
         Use depends on the type of the plotting function
     """
-    import matplotlib as mpl
+    if TYPE_CHECKING:
+        import matplotlib as mpl
+    else:
+        mpl = attempt_import("matplotlib")
 
     if isinstance(levels, Iterable):
         levels = sorted(levels)