Ocelot

Fixes a regression added in a previous release that prevented simulation of clusters observed with Gaia.

... remember: _always_ run your unit tests! 🥴

Two changes to how simulated clusters handle having zero stars were made.

- When zero stars are simulated, `ocelot.simulate.SimulatedCluster` now raises a `NotEnoughStarsError` (defined in `ocelot.simulate.errors`), making it easier to catch when no stars were simulated.
- When cluster pre-pruning causes zero stars to be simulated, `ocelot.simulate.SimulatedCluster` now also raises a `NotEnoughStarsError` (previously, this behaviour was undefined.)

A couple of improvements were made to cluster simulation:

- `ocelot.model.observation.BaseObservation` now has two additional default methods for _applying_ photometric and astrometric errors. These can be overwritten for observation types that require additional functionality beyond simple Gaussian errors.
- The `GaiaDR3ObservationModel` class now uses this photometric data overwrite to model underestimated BP and RP fluxes in Gaia DR3, slightly improving the realism of cluster CMDs further.


v0.4.2: Limit Python version

This is a quick release to limit our Python version to below 3.13, as it seems some of our dependencies don't support it yet.

The imf library has been moved to be an optional dependency to try and fix issues with PyPI upload failing.

This release brings a brand new, sophisticated API for cluster simulation, in addition to optimized models covering a number of different aspects of star cluster science. Additions include:

- The new `ocelot.simulate` submodule for simulating clusters, including:
- The flexible and hackable `SimulatedCluster` class. Extensive code optimizations allow for most open clusters to have full simulations (even randomly sampling the orbits of their binary stars to see if they would or wouldn't be resolved!) in less than a second
- New `SimulatedClusterParamters`, `SimulatedClusterModels`, and `SimulatedClusterFeatures` dataclasses for controlling `SimulatedCluster` functionality
- Ability to simulate observations of simulated clusters, allowing the same simulated cluster to be 'observed' by many different telescopes
- The new `ocelot.model` submodule for star cluster models, including:
- A model of King 1962 empirical star clusters. Currently limited to sampling clusters in 3D
- A bespoke model of differential extinction that approximates dust structure with fractal noise
- A heavily optimized implementation of the Moe & DiStefano 2017 binary star models
- Models for selection functions due to observations, such as an optimized implementation of the Castro-Ginard 2023 subsample selection function that can be used to model any generic subsample of stars
- A model for star cluster observations with Gaia
- Extensive unit tests to assure the reliability and accuracy of simulated clusters and their models
- Documentation improvements, including the first tutorial in the module
- Refactoring of some old aspects of the module

The APIs of `ocelot.simulate` and `ocelot.model` are now the first stable APIs of ocelot, and are ready to use in production code.