Backport ITP bugfixes from DiffEqBase in high precision contexts

[DaniGlez]

Fix a couple of issues on ITP that we found on SciML/DiffEqBase.jl#917
Specifically:

• Prevent iterations with zero progress (and thus getting stuck in a value until max iters) due to loss of relative precision near the solution
• Allow for reversed tspans

I also extended the bracketing tests to cover cases of IntervalNonlinearProblem with a reversed tspan (ie tspan[2] < tspan[1]). I am not sure if this is intended behavior; I assume it is since it works for the previous methods and the corresponding problem can be created. If this is not intended, we can remove those tests but I think it should error at IntervalNonlinearProblem creation time, then.

Another point: return behaviour is slightly different from the DiffEqBase implementation, where the solution has to fulfill iszero(f(nextfloat(t_sol))) || f(t_sol) * f(nextfloat(tsol)) < 0.0. Here, if f(t) = 0.0 then t is returned as the solution, which I guess makes more sense for other purposes.

[codecov]

Codecov Report

Merging #74 (73a977b) into main (26b5799) will increase coverage by 0.61%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #74      +/-   ##
==========================================
+ Coverage   92.48%   93.09%   +0.61%     
==========================================
  Files          21       21              
  Lines        1011     1014       +3     
==========================================
+ Hits          935      944       +9     
+ Misses         76       70       -6     

Impacted Files	Coverage Δ
src/itp.jl	91.93% <100.00%> (-1.29%)	⬇️

... and 2 files with indirect coverage changes

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