Vectorise (and varmat) `log_diff_exp` and `log_sum_exp` binary overloads

[andrjohns]

Description

The binary overload for log_sum_exp currently only accepts two scalar arguments (i.e., not two matrices to operate elementwise on). While the binary overload for log_diff_exp does accept two matrix/container arguments, the vectorisation is handled by apply_scalar_binary.

For both of these functions there are likely to be performance benefits from adding vectorised & varmat implementations

While not directly related, the context is for optimising (Gaussian) copula models

Current Version:

v4.4.0