Add support for complex noise formulae in roadrunner

[PaulJonasJost]

This currently assumes two things:

• The noise formula for one observable is always the same
• There is no time dependent changes in the parameters of the formula

I think both assumptions are things we make implicitly in PEtab/pyPESTO?

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 84.39%. Comparing base (45edfd2) to head (761eb3e).

:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## develop #1363 +/- ## =========================================== - Coverage 84.44% 84.39% -0.06% =========================================== Files 157 157 Lines 12899 12914 +15 =========================================== + Hits 10893 10899 +6 - Misses 2006 2015 +9 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[dweindl]

• There is no time dependent changes in the parameters of the formula

I think both assumptions are things we make implicitly in PEtab/pyPESTO?

I don't think pypesto should assume that without verification. For PEtab/amici it's just a limitation of amici.