Use Brent's method instead of bisection for computing apsides and nodes

[pleroy]

This improves performance by a few percents. Before:

--------------------------------------------------------------------------
Benchmark                                Time             CPU   Iterations
--------------------------------------------------------------------------
ApsidesBenchmark/ComputeApsides  119451153 ns    115664830 ns          362
ApsidesBenchmark/ComputeNodes     36542126 ns     35692924 ns         1191

after:

--------------------------------------------------------------------------
Benchmark                                Time             CPU   Iterations
--------------------------------------------------------------------------
ApsidesBenchmark/ComputeApsides  113703789 ns    112403921 ns          375
ApsidesBenchmark/ComputeNodes     33949542 ns     33410740 ns         1235

This improvement was discussed in #2913.