OBBT issue fixes

[fatihcengil]

• Fixed OBBT average reduction issue
• Fixed bug in best_rel_gap evaluation
• Added new termination criteria to OBBT (max_reduction, max_width, rel_gap/iteration of OBBT)
• Fixed the issue to tighten only original nonlinear variables (aux vars were counted before)
• presolve_bt_bound_tol changed to 1E-4
• presolve_bt_obj_bound_tol added to avoid infeasibility in OBBT iterations

@harshangrjn Please let me know if any new tests are necessary for new OBBT criteria

[fatihcengil]

Issue #181 should be fixed

[codecov[bot]]

Codecov Report

Merging #198 (9abc4e9) into master (d17c69d) will increase coverage by 0.63%. The diff coverage is 90.76%.

:exclamation: Current head 9abc4e9 differs from pull request most recent head 5edee39. Consider uploading reports for the commit 5edee39 to get more accurate results

@@            Coverage Diff             @@
##           master     #198      +/-   ##
==========================================
+ Coverage   86.21%   86.85%   +0.63%     
==========================================
  Files          16       16              
  Lines        3164     3226      +62     
==========================================
+ Hits         2728     2802      +74     
+ Misses        436      424      -12     

Impacted Files	Coverage Δ
src/Alpine.jl	100.00% <ø> (ø)
src/multi.jl	74.72% <0.00%> (ø)
src/amp.jl	88.59% <69.23%> (+0.38%)	:arrow_up:
src/utility.jl	83.88% <81.81%> (+0.26%)	:arrow_up:
src/presolve.jl	85.80% <84.72%> (+4.40%)	:arrow_up:
src/MOI_wrapper/MOI_wrapper.jl	85.92% <92.72%> (+2.17%)	:arrow_up:
src/algorithm.jl	89.54% <100.00%> (+2.88%)	:arrow_up:
src/bounds.jl	84.36% <100.00%> (+0.47%)	:arrow_up:
src/heuristics.jl	54.38% <100.00%> (ø)
src/log.jl	91.73% <100.00%> (+1.65%)	:arrow_up:
... and 7 more

:mega: Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[harshangrjn]

176 fixed!

[harshangrjn]

133 Doesn't seem like an issue. Added a unit test for expr dereferencing of @NLexpression.

[harshangrjn]

108 fixed! Lower bounds do not cross and Alpine terminates with OBBT converging to a global opt. Although, convergence of first NLP is very slow.