shell args: bash has no formal parameter list -- both the ground-truth measurement and GitGalaxy's own counting were incomplete #1518

Summary

args_exact_match for shell measured 0/3 (0%) in docs/self_scan/tree_sitter_accuracy_history.csv -- the worst score in the whole table, which reads as "GitGalaxy's shell args regex is badly broken." It wasn't -- both the measurement and the engine had a real, fixable gap, but neither was actually broken in the way the raw percentage suggested.

Root cause 1 (measurement): ground truth only checked the declaration

Bash functions (foo() { ... }) have NO formal parameter-list syntax at all, permanently, by grammar -- the () are always empty. tests/tools/tree_sitter_accuracy_audit.py's _get_param_count only checked the declaration node for a formal parameter field, correctly found nothing, and reported real=0 for every bash function regardless of true arity. GitGalaxy's own args regex instead reads the BODY for positional-parameter references ($1, $2, "$@"), which is the only way arguments are ever expressed in bash -- comparing that against a ground truth that only ever looked at the (permanently empty) declaration was an unfair, misleading comparison.

Root cause 2 (engine): single match only ever saw the first positional reference

Confirmed via brew/brew::symlink_target_directory, which references BOTH $1 and $2 in its body but measured got=1 -- GitGalaxy's args-counting takes one .search() per function and stops at the first match, so every positional-parameter reference after the first was silently dropped.

Fix (implemented, see PR)

• tests/tools/tree_sitter_accuracy_audit.py: new _count_shell_real_positional_max, walking the function body (via tree-sitter's own real parse) for every simple_expansion/expansion node, returning the highest positional index referenced. $0 (script name, not an argument) is correctly excluded.
• gitgalaxy/core/detector.py: new _count_shell_positional_max + a new opt-in _args_findall_max_groups routing flag, so shell's args-counting scans the WHOLE block (not just the first match) and takes the max index referenced.
• gitgalaxy/standards/language_standards.py: shell's args regex wrapped in one outer capture group so the new routing flag can target it.

args_exact_match: 0/3 (0%) -> 3/3 (100%).

Why this isn't "GitGalaxy loses to tree-sitter here"

For this one specific signal (parameter count), in this one specific circumstance (no formal parameter-list syntax exists at all), a plain AST/tree-sitter-only tool has nothing to read at the declaration and can only ever report 0 -- correct, but useless for coupling/complexity scoring. GitGalaxy's body-aware regex is the more informative measurement here. Full writeup: docs/why_gitgalaxy_beats_ast_here.md.

This only affects extraction-pillar args-counting accuracy (epic #813) and the args_exact_match measurement (epic #1227/#1261), not risk scoring beyond the signal's own coupling number.