Infer root type from the provided function

[Kolaru]

This PR changes how the internal root type is determine:

Now it uses the return type of the function, regardless of the type of the initial region given. This should avoid unnecessary conversions.

[github-actions]

Benchmark Results

	master	80dd3e1...	master / 80dd3e1...
Dietmar-Ratz/Dietmar-Ratz 1/IntervalRootFinding.Krawczyk	2.11 ± 0.04 μs	0.269 ± 0.029 ms	0.00786 ± 0.00086
Dietmar-Ratz/Dietmar-Ratz 1/IntervalRootFinding.Newton	2.09 ± 0.05 μs	0.273 ± 0.033 ms	0.00766 ± 0.00094
Dietmar-Ratz/Dietmar-Ratz 2/IntervalRootFinding.Krawczyk	0.0989 ± 0.00066 ms	2.87 ± 0.049 ms	0.0345 ± 0.00063
Dietmar-Ratz/Dietmar-Ratz 2/IntervalRootFinding.Newton	0.234 ± 0.0099 ms	3.11 ± 0.069 ms	0.0754 ± 0.0036
Dietmar-Ratz/Dietmar-Ratz 3/IntervalRootFinding.Krawczyk	0.328 ± 0.011 ms	1.85 ± 0.073 ms	0.177 ± 0.0093
Dietmar-Ratz/Dietmar-Ratz 3/IntervalRootFinding.Newton	0.244 ± 0.011 ms	1.78 ± 0.072 ms	0.137 ± 0.0084
Dietmar-Ratz/Dietmar-Ratz 4/IntervalRootFinding.Krawczyk	0.0472 ± 0.00046 ms	3 ± 0.059 ms	0.0158 ± 0.00034
Dietmar-Ratz/Dietmar-Ratz 4/IntervalRootFinding.Newton	30.3 ± 0.38 μs	3 ± 0.068 ms	0.0101 ± 0.00026
Dietmar-Ratz/Dietmar-Ratz 5/IntervalRootFinding.Krawczyk	1.92 ± 0.05 μs	0.104 ± 0.019 ms	0.0186 ± 0.0035
Dietmar-Ratz/Dietmar-Ratz 5/IntervalRootFinding.Newton	1.94 ± 0.051 μs	0.104 ± 0.019 ms	0.0187 ± 0.0034
Dietmar-Ratz/Dietmar-Ratz 6/IntervalRootFinding.Krawczyk	11.6 ± 0.16 μs	2.99 ± 0.075 ms	0.00388 ± 0.00011
Dietmar-Ratz/Dietmar-Ratz 6/IntervalRootFinding.Newton	18.2 ± 0.22 μs	3 ± 0.076 ms	0.00609 ± 0.00017
Dietmar-Ratz/Dietmar-Ratz 7/IntervalRootFinding.Krawczyk	2.17 ± 0.051 μs	2.3 ± 0.31 ms	0.000947 ± 0.00013
Dietmar-Ratz/Dietmar-Ratz 7/IntervalRootFinding.Newton	2.21 ± 0.06 μs	2.3 ± 0.25 ms	0.000961 ± 0.00011
Dietmar-Ratz/Dietmar-Ratz 9/IntervalRootFinding.Krawczyk	2.43 ± 0.06 μs	0.095 ± 0.019 ms	0.0255 ± 0.005
Dietmar-Ratz/Dietmar-Ratz 9/IntervalRootFinding.Newton	2.46 ± 0.06 μs	0.0952 ± 0.017 ms	0.0258 ± 0.0047
Rastigrin stationary points/IntervalRootFinding.Krawczyk	0.148 ± 0.0011 s	0.168 ± 0.0014 s	0.884 ± 0.0097
Rastigrin stationary points/IntervalRootFinding.Newton	0.141 ± 0.0014 s	0.157 ± 0.0061 s	0.894 ± 0.036
Smiley/Smiley and Chun (2001), Example 2.2/IntervalRootFinding.Krawczyk	4.34 ± 0.063 ms	7.27 ± 1.1 ms	0.597 ± 0.089
Smiley/Smiley and Chun (2001), Example 2.2/IntervalRootFinding.Newton	3.84 ± 0.061 ms	6.66 ± 1 ms	0.577 ± 0.087
Smiley/Smiley and Chun (2001), Example 5.2/IntervalRootFinding.Krawczyk	0.0492 ± 0.0031 s	0.0569 ± 0.0079 s	0.864 ± 0.13
Smiley/Smiley and Chun (2001), Example 5.2/IntervalRootFinding.Newton	0.0441 ± 0.0029 s	0.0488 ± 0.0058 s	0.904 ± 0.12
Smiley/Smiley and Chun (2001), Example 5.4/IntervalRootFinding.Krawczyk	2.62 ± 0.039 ms	8.31 ± 1.1 ms	0.316 ± 0.042
Smiley/Smiley and Chun (2001), Example 5.4/IntervalRootFinding.Newton	2.98 ± 0.039 ms	8.58 ± 1.1 ms	0.347 ± 0.045
time_to_load	0.322 ± 0.005 s	0.399 ± 0.0016 s	0.807 ± 0.013

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

[Kolaru]

Benchmarks are slower due to the inference only, and fail because I had to change them for the future.

A longer benchmark (few seconds to a minute) would be nice to be less sensitive to that kind of short setup effects (see #142 ).