Normalize in multiline galerkin

[leburgel]

Normalizes the AC´ tensor after application of the effective Hamiltonian before using it to compute the galerkin. This seems more fair for stat mech problems with very large leading boundary eigenvalues, the current implementation was giving me quite a bit of trouble.

[codecov]

Codecov Report

Merging #78 (7a4d346) into master (a00ea68) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #78      +/-   ##
==========================================
+ Coverage   84.29%   84.33%   +0.03%     
==========================================
  Files          60       60              
  Lines        3726     3734       +8     
==========================================
+ Hits         3141     3149       +8     
  Misses        585      585              

Files	Coverage Δ
src/algorithms/toolbox.jl	94.15% <100.00%> (+0.32%)	⬆️

[lkdvos]

We should probably do the same for the other cases, while less likely, it is still possible to have a pathological Hamiltonian with energy close to zero, or very large energies.

Small note, in principle the left nullspace of AC is given by (1 - AL AL*) which is slightly faster. (See above). It might be a good idea to get all of these cases to be the same

[Jutho]

Doesn't normalize!(AC´) work instead of AC´/norm(AC´)? The latter gives one extra allocation. The former should work (and amount to scale!(AC´,1/norm(AC´))). If it doesn't, that's a bug.

[Jutho]

One less allocation could have been obtained by writing
out = mul!(AC´, state.AL[loc], state.AL[loc]' * AC´, -1, +1)