Add OpenDSS .dss parser

[samtalki]

Extend the parser to OpenDSS feeders. Distribution scope, opens the door to LinDist3Flow.

Scope

• New module powerio/src/format/opendss/ with a real lexer (token stream, not regex; .dss syntax has multi word commands, Redirect, ! comments).
• Public API: parse_opendss(content) -> Result<Network>, parse_opendss_file(path).
• Phase A: Circuit, Vsource, Line, Transformer, Load, Capacitor, Generator. Reactor / switch later.
• Map OpenDSS impedance matrices (Rmatrix, Xmatrix per phase) into Branch. Decide: keep balanced positive sequence in Network and add a Phase3Case for full 3 phase, or extend Branch with optional 3x3 phase impedance.

Reference cases

IEEE 13, 34, 37, 123 bus radial feeders (BSD licensed in OpenDSS distribution). Vendor under tests/data/opendss/.

Acceptance

• Lexer with unit tests for Redirect, New <element>.<name>, multi line commands, ! comments.
• IEEE-13 round trips through parse_opendss → build_ybus → solve and matches OpenDSS's reference Y_bus to 1e-9.
• At least one radial feeder is detected by IndexedNetwork::is_radial().

[samtalki]

Flagging a scope conflict rather than closing unilaterally. The README now states caseio "targets the transmission planning / OPF interchange formats" and is "deliberately not a CIM tool and not a distribution-feeder modeler." OpenDSS is a distribution-feeder format and the roadmap dropped it, so this issue is at odds with the current direction. The 3-phase Phase3Case sketch here is also out of step with the bus/branch/gen/load/shunt Network hub the converter is built on.

If distribution is still wanted, this needs reframing — e.g. a balanced positive-sequence OpenDSS reader producing Network, explicitly leaving full 3-phase to a CIM hub. Otherwise recommend closing as out of scope. Leaving open for a direction call.

[samtalki]

PR #82 implements this, reframed: the dss parser lands in the new powerio-dist crate as a multiconductor model in wire coordinates, not as a positive sequence mapping into the transmission Network. The discussion here had already leaned that way; collapsing a feeder to positive sequence loses exactly the data the distribution formats exist to carry. The lexer is the real token stream this issue asked for (TParser semantics from the OpenDSS source: quote pairs, ~/More continuations, Redirect/Compile, RPN, property name prefixes, | matrices, bus dotting), with per construct unit tests. Phase A covers Circuit/Vsource, Line, LineCode, Load, Transformer, Capacitor, Generator, switches via SwtControl, RegControl as fixed tap; the deferred classes parse into a raw layer and convert with explicit warnings (#84 tracks typed support).

Verification went beyond the Y_bus criterion here: every emitted .dss re-solves in OpenDSS (opendssdirect, controls off, tolerance 1e-10) and matches the original per node voltages to 1e-8 pu where conversion fidelity is the only variable, on IEEE 13/34/123 plus eight micro cases.

One finding worth surfacing on the tracker: the draft BMOPF schema's matrix key patterns (^R_series_\d_\d and friends in frederikgeth/bmopf-report) reject two digit conductor indices, so a 10x10 linecode emits R_series_10_10 keys the schema cannot validate; the pattern is also unanchored at the tail. ^R_series_\d+_\d+$ fixes both. The linecode_10x10 fixture in the PR is the concrete demonstration, and the PR body carries the full schema feedback list.

[samtalki]

Superseded by #53, #84, and draft PR #82. The old powerio/src/format/opendss plan no longer matches the powerio-dist architecture.