GetStateIndex: wrap idx so it broadcasts as scalar across timeseries

[ChrisRackauckas-Claude]

Note

Please ignore this PR until reviewed by @ChrisRackauckas.

Summary

(gsi::GetStateIndex)(::Timeseries, prob, i) (at `src/state_indexing.jl:43`) currently does:

```julia
getindex.(state_values(prob, i), gsi.idx)
```

This works when `gsi.idx` is a scalar Int (broadcasts as length-1 against the time axis), but errors when `gsi.idx` is array-valued — e.g. a `Vector{Int}` produced when the symbolic is array-typed like `x(t)[1:3]` — because the broadcast then tries to zip the length-N timeseries against the length-K index vector:

```
DimensionMismatch: arrays could not be broadcast to a common size:
a has axes Base.OneTo(N) and b has axes Base.OneTo(K)
```

The companion no-`i` overload (line 38) already handles this with a 1-tuple wrap:

```julia
function (gsi::GetStateIndex)(::Timeseries, prob)
return getindex.(state_values(prob), (gsi.idx,)) # wrapped
end
```

This PR mirrors that wrap on the with-`i` overload so array-valued `idx` is always broadcast as a scalar across the time axis.

Reproduction

Triggered by `sol[x(t)[1:3], :]` in the RecursiveArrayTools.jl downstream tests with MTK 11 + OrdinaryDiffEq 7. After fix, the call returns the expected `Vector{Vector{Float64}}` matching `sol[x(t)[1:3]]`.

Side note

The same issue is being worked around on the RecursiveArrayTools.jl side in SciML/RecursiveArrayTools.jl#594 by short-circuiting `A[sym, :] → A[sym]` for symbolic indices; once this SII fix lands and gets released, that workaround can be removed.

Test plan

• Existing `Pkg.test()` passes locally (modulo the pre-existing `jet_test.jl:161` `remake_buffer` failure that's unrelated to this change and reproduces on master).
• RAT downstream `DiffEqArray Indexing Tests` safetestset goes 357/358 → 358/358 when this fix is loaded.
• CI