ExaPowerIOExt — to_powerdata + parse_ac_power_data + parity test

[samtalki]

Weak-dependency extension on ExaPowerIO (uuid 14903efe-9500-4d7f-a589-7ab7e15da6de, the typed data layer ExaModelsPower consumes). to_powerdata reproduces parse_matpower_inner exactly (per-unit ÷baseMVA, radians, c1..c8 via the BranchData ctor, dense bus reindex, filtered PV/slack inference), producing an ExaPowerIO PowerData{T}; plus a drop-in parse_ac_power_data NamedTuple so PowerIO feeds ExaModelsPower.build_polar_opf / build_rect_opf / build_dcopf directly (there is no opf_model entry point — that was a mistake in the original issue).

Value-add: this brings the formats ExaPowerIO can't read (PSS/E, PowerWorld, PowerModels JSON) into the Exa optimization stack. DoD: value-for-value vs ExaPowerIO's filtered=true parse.

[samtalki]

Heads up for this extension: PowerIO.jl now exposes to_normalized(net::Network) at the binding level (#13). It returns the per-unit / radian / tap-0→1 / out-of-service+ISOLATED-filtered / dense-1-based-reindex view with PV/REF/PQ inference — the same normalization ExaPowerIO's filtered=true parse produces, computed once in the Rust core (pio_to_normalized). to_powerdata and parse_ac_power_data can build on this instead of re-deriving the normalization in Julia.

Leaving this open for the extension itself (to_powerdata → PowerData{T}, parse_ac_power_data NamedTuple) and the value-for-value parity test.