Handle the "maybe extend" behaviour of extendMatrix() more appropriately.

[pratyai]

Currently, if the given SDDM has a very small diagonal, the extension does not happen, but the RHS is augmented anyway. This causes a crash.

Example:

using Laplacians
using Statistics
using SparseArrays
using LinearAlgebra

n = 5;
A = Laplacians.uni_chimera(5, 1) * 5;  # heavyweight connected graph
L = Laplacians.lap(A);
M = L + sparse(I, n, n) * eps(Float64);  # with lightweight extra diagonal value
params = Laplacians.ApproxCholParams(:deg, 2, 2);
f = Laplacians.approxchol_sddm(M; params = params, verbose = false);
b = randn(n); b .-= mean(b);

@show f(b)  # this used to crash

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 81.81818% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 76.49%. Comparing base (9ec1c10) to head (f01326a).

Files	Patch %	Lines
src/solverInterface.jl	81.81%	2 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #46      +/-   ##
==========================================
- Coverage   76.51%   76.49%   -0.03%     
==========================================
  Files          37       37              
  Lines        5596     5599       +3     
==========================================
+ Hits         4282     4283       +1     
- Misses       1314     1316       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.