Support generic initial_state in simulate config

[jc-macdonald]

Summary

Generalize the SimulateCommand to support arbitrary-length initial state vectors instead of the hardcoded s0/i0/r0 convention.

Problem

The simulate CLI command hardcodes initial state extraction to exactly three SIR compartments (s0, i0, r0). This makes it impossible to use flepimop2 simulate with models that have more (or fewer) than 3 states — e.g., SIRHD (5 states) or vaccination-structured models (9+ states).

Solution

When a simulate spec includes an initial_state dict, its values are used directly as the state vector and its keys are excluded from the params dict. The legacy s0/i0/r0 fallback is preserved for backward compatibility.

New config usage:

simulate:
  demo:
    times: 0.0:1.0:200.0
    initial_state:
      S: 9990
      I: 10
      H: 0
      R: 0
      D: 0

No schema changes were needed — SimulateSpecificationModel already has ConfigDict(extra="allow"), so the initial_state key is accepted and accessible via getattr.

Note on inference compatibility

The new initial_state path currently takes raw floats, bypassing ParameterABC. This means initial conditions are fixed constants that cannot yet participate in sampling or optimization. However, this does not block future inference support — the dict-of-names structure is well-positioned for it. A future extension just needs to check whether each value is a bare number or a parameter module reference (e.g., {module: distribution, ...}) and route accordingly. The key names are the actual state names, which is a better foundation than the ad-hoc s0/i0/r0 convention.

Changes

• src/flepimop2/_cli/_simulate_command.py: Generalized initial state extraction (+22/-12 lines)
• CHANGELOG.md: Added entry under Changed

Testing

• All 291 existing tests pass (289 unit + 2 integration)
• ruff and mypy clean

[TimothyWillard]

Conceptual question, also this should be a squashed commit if merging

[TimothyWillard]

This is moving in a different direction than #211 by putting the initial conditions into the simulate config block rather than the system config block. Is that where we want to have initial conditions be long term? If so #211 probably requires reworking.

[jc-macdonald]

superseded by #226.