fixed typos in docs and docstring (#3231)

Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>

docs/source/guide/nn-construction.rst

@@ -35,7 +35,7 @@ The construction function performs the following steps:
 
 In construction function, one first needs to set the data dimensions. For
 general PyTorch module, the dimensions are usually input dimension,
-output dimension and hidden dimensions. For graph neural, the input
+output dimension and hidden dimensions. For graph neural networks, the input
 dimension can be split into source node dimension and destination node
 dimension.
 

docs/source/guide/training-edge.rst

@@ -156,7 +156,7 @@ You can similarly write a ``HeteroMLPPredictor``.
 
 .. code:: python
 
-    class MLPPredictor(nn.Module):
+    class HeteroMLPPredictor(nn.Module):
         def __init__(self, in_features, out_classes):
             super().__init__()
             self.W = nn.Linear(in_features * 2, out_classes)

docs/source/guide/training-graph.rst

@@ -101,7 +101,7 @@ where :math:`h_g` is the representation of :math:`g`, :math:`\mathcal{V}` is
 the set of nodes in :math:`g`, :math:`h_v` is the feature of node :math:`v`.
 
 DGL provides built-in support for common readout operations. For example,
-:func:`dgl.readout_nodes` implements the above readout operation.
+:func:`dgl.mean_nodes` implements the above readout operation.
 
 Once :math:`h_g` is available, one can pass it through an MLP layer for
 classification output.

python/dgl/nn/pytorch/utils.py

@@ -108,7 +108,7 @@ class Sequential(nn.Sequential):
 
     Description
     -----------
-    A squential container for stacking graph neural network modules.
+    A sequential container for stacking graph neural network modules.
 
     DGL supports two modes: sequentially apply GNN modules on 1) the same graph or
     2) a list of given graphs. In the second case, the number of graphs equals the