Adds Evidence estimator module

[maho3]

Adds implementations of evidence estimators from:

• Harmonic (McEwen et al. 2021, Spurio-Mancini et al. 2023)
• Evidence Networks (Jeffrey & Wandelt 2023)

Implements runners for each of these methods, and compares their ability to measure $\log K$ in notebooks/evidence.ipynb.

Still to-do:

• [ ] Add capacity for embedding nets
• [ ] Add capacity for ensembling
• [ ] Add functional documentation
• [ ] Add tests

[codecov-commenter]

:warning: Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 0.46948% with 212 lines in your changes missing coverage. Please review.

Project coverage is 85.88%. Comparing base (3019bbd) to head (4733d3b).

Files	Patch %	Lines
ili/evidence/runner.py	0.00%	144 Missing :warning:
ili/evidence/utils.py	0.00%	67 Missing :warning:
ili/evidence/__init__.py	0.00%	1 Missing :warning:

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

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #157 +/- ## =========================================== - Coverage 96.95% 85.88% -11.08% =========================================== Files 23 26 +3 Lines 1644 1856 +212 =========================================== Hits 1594 1594 - Misses 50 262 +212 ```

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

[DeaglanBartlett]

As highlighted in https://github.com/astro-informatics/harmonic/issues/271, using the harmonic code for models with only one parameter does not work, since the normalising flow expects there to be at least two parameters. Even if this is not fixed in harmonic, we should create a work around here. Creating a simple normalising flow which we then wrap to become a hm.model.FlowModel object so that it has the correct functionality should be sufficient. This edge case should be explicitly included in the unit tests.