Add pinned numpy and fix build (pytorch#155129)

Not sure why the online doc build passes but it fails locally with these broken strings...

Also pinning numpy version even though it is technically optional to ensure users have the right version as most users have numpy in their environment anyways.
Pull Request resolved: pytorch#155129
Approved by: https://github.com/janeyx99, https://github.com/svekars

Author: albanD

.ci/docker/requirements-docs.txt

@@ -58,3 +58,5 @@ sphinx-copybutton==0.5.0
 sphinx-design==0.4.0
 sphinxcontrib-mermaid==1.0.0
 myst-parser==0.18.1
+
+numpy==1.26

torch/nn/attention/bias.py

@@ -36,14 +36,14 @@ class CausalVariant(IntEnum):
 
     Defines two types of causal biases:
 
-    `UPPER_LEFT`: Represents upper-left triangular bias for standard causal attention.
+    ``UPPER_LEFT``: Represents upper-left triangular bias for standard causal attention.
     The equivalent pytorch code for constructing this bias is:
 
     .. code-block:: python
 
         torch.tril(torch.ones(size, dtype=torch.bool))
 
-    For instance, with `shape=(3,4)`, the materialized bias tensor will be:
+    For instance, with ``shape=(3,4)``, the materialized bias tensor will be:
 
     .. code-block:: text
 
@@ -52,7 +52,7 @@ class CausalVariant(IntEnum):
          [1, 1, 1, 0]]
 
 
-    `LOWER_RIGHT`: Represents lower-right triangular bias, the include values are aligned to the lower
+    ``LOWER_RIGHT``: Represents lower-right triangular bias, the include values are aligned to the lower
     right corner of the matrix.
 
     The equivalent pytorch code for constructing this bias is:
@@ -65,7 +65,7 @@ class CausalVariant(IntEnum):
             diagonal=diagonal_offset,
         )
 
-    For instance, with `shape=(3,4)`, the materialized bias tensor will be:
+    For instance, with ``shape=(3,4)``, the materialized bias tensor will be:
 
     .. code-block:: text