Parse PSS/E v34 RAW files and tolerate two record shapes that aborted the build

[annacasavant]

v34 parsing

• Revision is read from the REV header field. The @! column marker no longer implies v35; v34 exports carry it too.
• One (sections, dtypes, defaults) registry per layout family. v34 reuses the v35 layout with v33 field order in seven records, new fields appended.
• Stage 2 branches on version families instead of literal revision strings.
• Fixes found along the way: double-quoted strings kept their quotes, and the v35 VSC subline table was one field short.

Duplicate two-winding transformers

• Records on the same bus pair with the same circuit id collided in the id registry. The first record is kept, later duplicates are dropped with a warning naming the PSS/E record.

Out-of-service VSC lines

• The reader errored when no converter was TYPE 1, the shape PSS/E writes for MDC = 0. The check is now gated on the in-service flag; such lines are kept as unavailable with placeholder setpoints and a warning.

Verification

• Full suite passes, 2,563 assertions, including new fixtures for each change.

[codecov]

Codecov Report

❌ Patch coverage is 90.26549% with 11 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/pm_io/pti.jl	89.06%	7 Missing ⚠️
src/pm_io/psse.jl	91.83%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!

[m-bossart]

Proposal: reject duplicate TRANSFORMER records in the pti → PowerModels translation

Move the duplicate-transformer rejection out of the openapi layer and into
_psse2pm_transformer! (src/pm_io/psse.jl).

Why there

• pti stays a faithful record reader.
• _psse2pm_transformer! iterates pti_data["TRANSFORMER"], a Vector in file order, so
  "keep the first record" is guaranteed by construction rather than by sorting on pm
  index downstream.
• Every consumer of parse_file / PowerModelsData gets the PSS/E behavior, not just the
  openapi build.
• The rule is keyed on the PSS/E identity, not on a display name that
  branch_name_formatter can change.

Changes in _psse2pm_transformer!

1. Before the loop, create a Set of seen identities.
2. For each record, build the key (I, J, K, uppercase(strip(CKT))).
   This one key covers two-winding (K == 0) and three-winding records.
   It uses the RAW bus numbers, not the routed node-bus endpoints.
3. If the key was already seen, @warn with the buses and CKT, then continue.
   Otherwise record it and build the entry as today.
4. Write the stripped CKT into the two-winding source_id (currently the raw
   pop!(transformer, "CKT")).
   Three-winding already strips it for "circuit".

Changes in openapi

• Remove _is_duplicate_branch and its call in read_branches!
  (src/openapi/branch.jl).
• register! goes back to being the only guard.
  A duplicate name reaching it then means a parser bug, and it fails loudly.

Tests

• Add a parse-level test in test/test_parse_psse.jl using
  synthetic_v35_duplicate_transformer_names.raw.
  It checks the warning, a single 201–202 transformer entry holding the first record's
  values (x = 0.05, rating 40), and that the 203–204 transformer is untouched.
• Add a record whose CKT differs only in whitespace (e.g. ' 1' vs '1 ') to the
  fixture, or to a new fixture.
  Both current duplicate records use '1 ', so the fixture doesn't cover the real case.
• Keep the openapi test in test/test_openapi_branch.jl as a build test, but drop the
  warning expectation.
  It then checks that the build succeeds and that one circuit survives.