Fix ITP for problems with flipped sign

[DaniGlez]

For some problems, there's a potential bug as the current ITP implementation follows the pseudocode in the Oliveira - Takahashi paper, which assumes f(left) < 0 < f(right), and thus it may wrongly contract the interval for problems with the reverse sign configuration (surprisingly enough, it sometimes works with the wrong signs [1]). This PR fixes the issue and adds a basic test for it.

[1] Did not realize that's only the case if one of the edges is the solution already

[ChrisRackauckas]

Thanks! And that test will be super helpful as well.

[codecov]

Codecov Report

Merging #72 (fe74b25) into main (8717ca6) will increase coverage by 0.70%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #72      +/-   ##
==========================================
+ Coverage   91.78%   92.48%   +0.70%     
==========================================
  Files          21       21              
  Lines        1010     1011       +1     
==========================================
+ Hits          927      935       +8     
+ Misses         83       76       -7     

Impacted Files	Coverage Δ
src/itp.jl	93.22% <100.00%> (+3.56%)	⬆️

... and 5 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[yash2798]

Thanks @DaniGlez for the fix :)