Refactor status effect system

[paulcscharf]

Based on #421

This:

• Minimizes allocations (only the scope and the list of effects are allocated, everything else lives on the stack)
• Minimizes boilerplate (primarily by removing the phenomenon types)
• Improve performance (and prevent boxing) by requiring IEquatable (which records implement by default)
• Simplifies scope life cycle further and fixes bug, as suggested in 📚 Break down the status effect state machine into smaller steps #421

[tomrijnbeek]

I only reviewed the system files, not the implementations. I assume they will change anyway if any of the feedback changes the systems.

A bit of general feedback on the shape of this PR. This PR is big, and ultimately it's trying to do three separate things all at once:

1. Allocation optimisation
2. Architectural changes
3. Style changes

The fact that this is all in one big piece makes it a bit hard to judge on. The allocation optimisation is largely non-controversial and if those had been done in a PR in isolation, would likely have been merged with little comment. The architectural changes are a bit more controversial, and are really the source of the biggest comments I had. The style choices - merging partial classes into a single file - are not really necessary. Especially since this is a PR to a PR, I feel not making stylistic changes helps avoid making the PRs more controversial than they already are.

I understand the argument for merging things into a single PR, but it would've been nice to see the work at least split out in commits, so that:

1. It's easier to review the different changes in isolation
2. It's easier to cherrypick the changes we want if we agree on the non-controversial stuff but disagree on the controversial topics