[WIP] A bunch of updates to allow for negative-valued tensor trains and improve robustness against under/over flow

[stecrotti]

I'm #32, but stronger.

This normalizes intermediate results of orthonormalize_right!, orthonormalize_left to prevent numerical issues.
Plus, it incorporates a "normalization" constant z directly in the tensor train object. Pseudo-code summary:

struct TensorTrain
   tensors
   z
end

evaluate(A::TensorTrain, x) = tr(prod(AA(xx) for (AA,xx) in zip(A,x))) / A.z
normalization(A::TensorTrain) = (l, z = accumulate_L(A); z / A.z)

Any operation which, in order to prevent numerical issues, needs to re-scale the tensors, must update A.z accordingly.
At the same time, any function which computes properties (evaluate, normalization, etc) must take into account the value of A.z

Fixes #31 #34 #35

[codecov]

Codecov Report

Attention: Patch coverage is 99.15254% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 96.01%. Comparing base (7fe8f80) to head (820a7ad).
Report is 2 commits behind head on main.

Files	Patch %	Lines
src/utils.jl	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #36      +/-   ##
==========================================
+ Coverage   95.13%   96.01%   +0.87%     
==========================================
  Files           6        6              
  Lines         329      376      +47     
==========================================
+ Hits          313      361      +48     
+ Misses         16       15       -1     

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