Added Typechecking to input tensor in RNN (#100100)
The input tensor of the RNN forward must be the same type as the weights.
While passing tensor of type long the error is:
`RuntimeError: expected scalar type Long but found Float`
Which is misleading because it said to convert Something to Long, but the correct solution is to convert the input to Float (Which is the type of the weights).
The new error:
`RuntimeError: input must have the type torch.float32, got type torch.int64`
Is correct and more verbose
Fixes #99998
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100100
Approved by: https://github.com/drisspg
