v0.7.2

• CLI case discovery is recursive and covers every supported format (#260):
  powerio batch and the TUI walk the input directory for .m, .raw,
  .aux, .epc, .pwb, .json, and .dss files, matched case
  insensitively, pruning hidden directories and the output directory. A file
  that fails to load during a scan is skipped with a warning instead of
  aborting the run, and the no-cases error names the supported extensions.
  Distribution cases (.dss, BMOPF/PMD JSON) parse to the multiconductor
  model and go through the explicit lower_multiconductor_to_balanced pass,
  whose approximations and dropped fields surface as warnings.
• A leading UTF-8 byte order mark no longer defeats the JSON classifier or
  any text reader. classify_json_text, the transmission read funnel, the
  distribution parsers (including .dss and redirected files), the PMD/BMOPF
  JSON split, Network::from_json, and Package::from_json all strip it,
  and the parse warnings itemize the removal (a same-format echo returns the
  text without the mark).
• parse_str_with_name: parse_str plus the name hint role the file stem
  plays in parse_file. The CLI now reads and classifies a .json case
  once and hands the text straight to the typed parser; a batch scan
  previously read each .json twice and DOM-parsed it three times.
• A nameless distribution case loaded by the CLI takes its file stem as the
  network name, so batch exports no longer collide on the lowering's
  lowered-multiconductor fallback.
• JSON reader hardening from a review pass over the family:
  • PowerModels: a status field written as a JSON boolean ("br_status": false) reads out of service instead of in service; baseMVA must be
    finite and positive; gencost rows padded to the MATPOWER matrix width are
    trimmed to the declared ncost before the per-unit unscale; an
    out-of-range cost model number passes through unscaled instead of
    wrapping into the piecewise/polynomial rescale.
  • pandapower: a trafo sn_mva of zero or below falls back to the system
    base instead of dividing the impedance by zero; string cells can no
    longer smuggle "inf"/"nan" into numeric columns.
  • egret: a polynomial cost exponent key is bounded, so a few bytes of JSON
    can no longer demand an arbitrarily large allocation or index out of
    bounds.
  • GOC3: a duplicate simple_dispatchable_device uid is rejected instead of
    silently taking another device's time series bounds and cost.
  • Surge: a float bus reference too large for an index is rejected like the
    equivalent integer instead of saturating to usize::MAX.
  • BMOPF: matrix key indices and winding counts are bounded (a crafted
    document could previously demand gigabytes or quadratic work), a missing
    line length warns instead of propagating NaN silently, and the meta
    block is kept in extras instead of dropped.
  • PMD: a data_model other than ENGINEERING (the index based MATHEMATICAL
    model) is rejected instead of being misread as ENGINEERING.
  • .pio.json: a semver build tag containing a hyphen (1.0.0+build-x) no
    longer fails the schema version check.

What's Changed

• cli: recursive case discovery; JSON parser hardening; 0.7.2 prep by @samtalki in #260

Full Changelog: v0.7.1...v0.7.2