Extract shared native argument fragment classification

Problem

PR #68 added equivalent native-argument fragment walks to the Bash and PowerShell parsers. Both implementations now collect contiguous lexer fragments, preserve the complete raw span and decoded value, detect opaque fragments, and safe-fail resolver-sensitive mixed quoting. Keeping those rules duplicated creates a drift risk when another fragment shape or shell is added.

This is not a request for a shared lexer or parser core. Bash and PowerShell must retain their shell-specific tokenization, quoting, recursion, and command semantics.

Proposed scope

Extract an internal shell-neutral classification result for one native argument assembled from parser-owned lexer fragments. Each parser adapts its own tokens into the helper and remains responsible for consuming the resulting token range.

The shared result should carry:

• the complete authored source span and raw value;
• the combined decoded logical value;
• whether any fragment is opaque or computed;
• whether all value fragments are literal;
• whether mixed literal and expandable fragments contain resolver-sensitive syntax;
• the first index after the consumed fragment run.

Non-goals

• Unifying the Bash and PowerShell lexers.
• Building a shared parser base class.
• Changing the public AST or parser APIs.
• Weakening DynamicSkip behavior for ambiguous mixed quoting.

Acceptance criteria

• Bash and PowerShell use one internal fragment-classification implementation.
• Shell-specific token adapters remain small and explicit.
• Existing Raw, Value, source-span, path, and DynamicSkip behavior remains unchanged.
• The PR #68 adversarial corpus cases for quoted values, maximal fragment runs, unquoted prefixes, mixed literal syntax, and curl @ transformation remain green in both shells.
• dotnet build -c Release, dotnet test -c Release, and header verification pass.

Extraction trigger

Do this before adding a third shell or another native fragment rule. Until then, the current duplication is correct but carries maintenance cost.

Comparez les parcours des fragments d’arguments natifs de Bash et de PowerShell introduits par PR #68, puis examinez leurs cas de corpus adversariaux pour les valeurs entre guillemets, les séquences maximales, la syntaxe mixte et la transformation de curl @. N’extrayez que la classification indépendante du shell, en gardant séparés les adaptateurs de tokens et le comportement du parseur. Exécutez dotnet build -c Release, dotnet test -c Release et la vérification des headers ; c’est terminé lorsque tout le comportement existant reste au vert dans les deux shells.