feat: add coord_shift transition type

[jc-macdonald]

Closes #47

Summary

Introduces coord_shift entries in the transitions list as a concise way to specify axis-coordinate-shifting transitions across a set of states.

What changed

• New transition type coord_shift: instead of writing individual from/to transitions for every state and every non-shifted axis combination, a single coord_shift entry expands to all concrete transitions automatically.
• Expansion logic runs at the same pipeline stage as chain expansion — mutating transitions_raw before template expansion — so the generated transitions are treated identically to any hand-written concrete transition.
• Validation covers: axis must exist, both coordinates must be valid members of that axis, apply_to must be a non-empty list, rate must be a string, all target states must already exist in the expanded state list.

Example

transitions:
  - coord_shift: {vax: u -> v}
    apply_to: [S, R]
    rate: nu

Expands to:

- {from: S__vax_u, to: S__vax_v, rate: nu}
- {from: R__vax_u, to: R__vax_v, rate: nu}

For states with additional axes, transitions are emitted for each combination of the non-shifted axis coordinates.

Tests

8 new tests:

• Single-axis expansion (correct pairs, correct count)
• Multi-axis expansion over non-shifted coordinates
• Coexistence with regular and templated transitions
• Validation errors: unknown axis, bad coordinate, missing apply_to, missing rate, malformed arrow syntax

[pearsonca]

Seems conceptually fine, but: what's the plan for expressing non-categorical shifts?

[jc-macdonald]

Seems conceptually fine, but: what's the plan for expressing non-categorical shifts?

for continuous shifts the natural thing is an advection operator - I have an issue for that already on op_engine, ACCIDDA/op_engine#59

[pearsonca]

Seems fine for now - I do think we want to have conceptual consistency in the configuration DSL expression for change in e.g. vaccine status and change in e.g. age, whatever is going on under the hood.

break-break, we'll want some simplified expression for initial conditions by axes as well