Matchbox recommender message 7

Hi everyone,

**I've been looking for where message 7 (m_⁎->s as labelled in the [Matchbox paper](https://www.microsoft.com/en-us/research/publication/matchbox-large-scale-bayesian-recommendations/) [1] factor graph) is implemented in the infer.net code**, but I can't seem to find it. As a study excercise, me and @glandfried have been trying to implement the collaborative filtering version of the Matchbox recommender algorithm, as described in the Matchbox paper. We have been able to implement almost everything. However, **our implementation of approximate message 7 differs greatly from the exact version of the message**, the bimodal one. In contrast, we had no issues with the approximate message 2.

If anyone could help me find the infer.net implementation of the message so we can compare I would appreciate it. So far I could only find approximate message 2 (m_⁎->z) at
[`infer/src/Runtime/Factors/Product.cs#ProductAverageLogarithm`](https://github.com/dotnet/infer/blob/main/src/Runtime/Factors/Product.cs#L2003).

As a reminder, I copy here the factor graph,

and the approximation equations (approximate messages 2 and 7 respectively),

From the original Matchbox paper. I interpret this as,

(μ_t and σ²_t denote the mean and variance of the (Gaussian) marginal distribution p(t). μ_{z ->⁎} σ²_{z ->⁎} denote the mean and variance of message 6).

**It would also be nice to get a hint to derive the approximations for 2 and 7 on our own (or a reference).**

Thanks in advance!

[1] Matchbox Paper: https://www.microsoft.com/en-us/research/publication/matchbox-large-scale-bayesian-recommendations/