fix(parser): preserve opam metadata after inline descriptions

[mstykow]

Summary

• Preserve later opam metadata when description: is inline or moves to the next quoted line, while keeping existing triple-quoted fixture behavior stable.
• Verify row 35 (OCaml / opam) with compare-outputs --profile common on ocaml/dune, ocaml/ocaml-lsp, and ocaml/merlin, then record the benchmark rows and mark the scorecard row verified.
• Compare artifacts: .provenant/compare-runs/20260422T115905Z-dune-32635/, .provenant/compare-runs/20260422T120550Z-ocaml-lsp-41966/, .provenant/compare-runs/20260422T120950Z-merlin-47624/.

Issues

• Covers: parser verification scorecard row 35 (OCaml / opam)

Scope and exclusions

• Included: parser regression tests for inline and next-line opam descriptions, opam benchmark rows, regenerated benchmark chart/stats, and the scorecard status update.
• Explicit exclusions: no unrelated parser refactors, no target-specific scan tuning, and no expected-output fixture rewrites.

Intentional differences from Python

• Treat .for-opam Dune workspace files as non-opam inputs even when ScanCode heuristically reports opam package data on them, because the file contents are Dune syntax rather than opam manifest syntax.
• Keep Unicode-preserving copyright and author normalization when it is more faithful than ScanCode's ASCII-degraded output.

Follow-up work

• Created or intentionally deferred: none.

Expected-output fixture changes

• Files changed: none.
• Why the new expected output is correct: the parser fix is covered by new regression tests plus the existing opam golden suite, which passed without updating any expected files.