DECTRIS/Merlin: tweak tile shape adjustment

Don't change the max I/O size, and be sure to select a full-frame
tileshape via the base shape. This should be more robust, even if it can
lead to non-optimal tiles (especially in the GPU case), which has to be
properly fixed in a different way.

src/libertem_live/detectors/dectris/acquisition.py

@@ -299,18 +299,6 @@ def need_decode(self, read_dtype, roi, corrections):
         ''
         return True
 
-    def adjust_tileshape(self, tileshape, roi):
-        ''
-        depth = 12  # FIXME: hardcoded, hmm...
-        return (depth, *self.meta.shape.sig)
-        # return Shape((self._end_idx - self._start_idx, 256, 256), sig_dims=2)
-
-    def get_max_io_size(self):
-        ''
-        # return 12*256*256*8
-        # FIXME magic numbers?
-        return 12*np.prod(self.meta.shape.sig)*8
-
     def get_base_shape(self, roi):
         ''
         return (1, *self.meta.shape.sig)

src/libertem_live/detectors/merlin/acquisition.py

@@ -176,19 +176,11 @@ def check_valid(self):
 
     def need_decode(self, read_dtype, roi, corrections):
         ''
-        return True  # FIXME: we just do this to get a large tile size
-
-    def get_max_io_size(self):
-        ''
-        # return 12*256*256*8
-        # FIXME magic numbers?
-        return 128 * np.prod(self.meta.shape.sig) * 8
+        return True
 
     def get_base_shape(self, roi):
-        return (1, 1, self.meta.shape.sig[-1])
-
-    def adjust_tileshape(self, tile_shape, roi):
-        return (tuple(tile_shape)[0], *self.meta.shape.sig)
+        ''
+        return (1, *self.meta.shape.sig)
 
     def get_partitions(self):
         ''