openbim-step 0.11.0

Added

• ParseOptions::accept_real_without_point: a data-section real written
  without the decimal point ISO 10303-21 requires before its exponent
  (1E-05, -2E3, 3e+2, +4E0) is read as the real it means, with the
  point inserted (1.E-05), and reported as one
  DiagnosticKind::RealWithoutPoint warning over the number's bytes. The
  stored value carries the point, so the writer emits valid Part 21.
  Incomplete numbers (1E, 1E-, 1EE2, .5E2) stay errors, strings are
  never touched, and the header stays strict. buildingSMART's own IFC4.x
  alignment test files write such reals.
• StepError::is_real_without_point: strict parsing refuses such a number
  with this typed error, its span covering exactly the number and its
  detail quoting it. Before, the error was a generic syntax error whose
  span ended at the E.
• Lazy decoding under options: decode_record_with,
  decode_record_borrowed_with and Scan::decode_with return the record
  exactly as parse_with reads it with the same options, plus that
  record's diagnostics. decode_record and decode_record_borrowed stay
  strict.

Changed

• ParseOptions::lenient() also accepts reals without a decimal point: a
  lenient parse now keeps such a record with a warning instead of skipping
  it. ParseOptions::strict().on_malformed_record(OnMalformed::Skip) keeps
  the old behaviour.

The common number path is unchanged; the new handling is out of line.
Measured on a 109 MB Revit IFC and a 102 MB STEP file against 0.10.0:
+0.3% and +0.6% user-space instructions for a strict parse.