update symeig backward note about similar eigenvalues (pytorch#52311) (pytorch#52446)

Summary:
First part of pytorch#49886 to at least properly warn users of the current state

Pull Request resolved: pytorch#52311

Reviewed By: soulitzer

Differential Revision: D26495644

Pulled By: albanD

fbshipit-source-id: 72abdfe41cdbcc1ac739a536eb85d1aa4ba90897

Author: albanD

torch/_torch_docs.py

@@ -8515,9 +8515,9 @@ def merge_dicts(*dicts):
 .. note:: Irrespective of the original strides, the returned matrix `V` will
           be transposed, i.e. with strides `V.contiguous().transpose(-1, -2).stride()`.
 
-.. note:: Extra care needs to be taken when backward through outputs. Such
-          operation is really only stable when all eigenvalues are distinct.
-          Otherwise, ``NaN`` can appear as the gradients are not properly defined.
+.. warning:: Extra care needs to be taken when backward through outputs. Such
+             operation is only stable when all eigenvalues are distinct and becomes
+             less stable the smaller :math:`\min_{i \neq j} |\lambda_i - \lambda_j|` is.
 
 Args:
     input (Tensor): the input tensor of size :math:`(*, n, n)` where `*` is zero or more