Add MvNormalMeanScalePrecision node

[albertpod]

This PR adds MvNormalMeanScalePrecision that implements a multivariate normal distribution node with a "scaled" diagonal precision matrix, i.e.
N(y|x, γI) where γ ~ Gamma(a, b) and I is the identity matrix.

The main difference between this node and the MvNormalMeanPrecision is the update equation for the precision parameter. Unlike MvNormalMeanPrecision, which sends a message proportional to Wishart, the MvNormalMeanScalePrecision will send a Gamma-like message.

When defining the model, I expect the user to add this node in the following way, i.e.

...
τ ~ Gamma(1.0, 1.0) # prior
y ~ MvNormal(μ=ones(2), γ=τ)
...

[codecov-commenter]

Codecov Report

Merging #184 (01a4b51) into master (c7d269c) will increase coverage by 0.13%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #184      +/-   ##
==========================================
+ Coverage   76.68%   76.81%   +0.13%     
==========================================
  Files         196      200       +4     
  Lines        6621     6660      +39     
==========================================
+ Hits         5077     5116      +39     
  Misses       1544     1544              

Impacted Files	Coverage Δ
src/ReactiveMP.jl	66.66% <ø> (ø)
src/nodes/mv_normal_mean_scale_precision.jl	100.00% <100.00%> (ø)
src/rules/mv_normal_mean_scale_precision/mean.jl	100.00% <100.00%> (ø)
src/rules/mv_normal_mean_scale_precision/out.jl	100.00% <100.00%> (ø)
.../rules/mv_normal_mean_scale_precision/precision.jl	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[bvdmitri]

Great @albertpod , thanks!