Speed up of cross_section 💨

[StFroese]

This speeds up calls to cross_section by a factor of 3 😮

There is some slow down of the cross section calculation described in #244 .

This reduces the number of internal calls of cross_section due to _temporary_kinematics context manager.

Before PR:

In [2]: from chromo.kinematics import FixedTarget
   ...: from chromo.models import Sibyll23e

In [3]: kin_init = FixedTarget(100, 2212, 'O16')
   ...: kin = FixedTarget(80, 2212, 'O16')
   ...:
   ...:
   ...: sib = Sibyll23e(kin_init)

In [4]: def cs(mod, kin):
   ...:     return mod.cross_section(kin).prod
   ...:
In [5]: %timeit cs(sib, kin)
992 μs ± 252 ns per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

With this PR:

In [8]: %timeit cs(sib, kin)
328 μs ± 791 ns per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

[StFroese]

needs a rebase after #243

[StFroese]

tests passed locally ✅ should pass here too, let's see

@afedynitch I think if we merge this after #243 I should do a patch release v0.10.1 because factor 3 speed up should be made available to users soon

[afedynitch]

Oh, and don't forget to update the PR description with the correct timings