Port ls argument surface from uutils via codegen

[chaliy]

Background

crates/bashkit/src/builtins/ls.rs is 3,605 LoC of handwritten flag parsing — the largest single GNU-parity gap in bashkit's coreutils port table. uu_ls's uu_app() is also the largest clap definition in uutils, so this port is the natural pressure test for the codegen tool: if it survives ls, the rest of coreutils is mechanical.

The PR also forces useful upgrades to bashkit-coreutils-port if uu_ls's en-US.ftl uses Fluent placeables or selectors that the current rewriter does not yet handle.

Acceptance criteria

• crates/bashkit/src/builtins/generated/ls_args.rs produced by cargo run -p bashkit-coreutils-port -- /tmp/uutils ls <REV>. If codegen aborts with unresolved translate!() for Fluent placeables/selectors, upgrade the codegen tool to handle the cases uu_ls exercises before committing the output.
• pub mod ls_args; added to crates/bashkit/src/builtins/generated/mod.rs.
• crates/bashkit/src/builtins/ls.rs switches its argument parser to ls_command().try_get_matches_from(...). Existing rendering logic stays put; the change is parser-only at this issue's scope.
• Flags accepted by uu_ls but not yet implemented by bashkit return an explicit "not yet implemented" error matching the existing tac -b/-r/-s pattern (no silent no-op).
• GNU help layout — Usage: first — via the same .help_template(...) chain cat.rs and textrev.rs apply.
• tests/spec_cases/bash/help-flag.test.sh ls_help continues to pass; ls_version updated if the version line format diverges.
• Spec tests in tests/spec_cases/bash/ cover: at least three short flags from uu_ls that bashkit previously parsed loosely (e.g. -Q, --quoting-style=, --group-directories-first), one composite scenario, and unknown-flag rejection (with ### bash_diff for the clap-vs-GNU exit-code difference).
• coreutils-args-drift.yml picks up ls_args automatically via the existing pub mod ... grep — no workflow edit required.
• Every existing ls spec test stays green.
• Compiled binary size delta on cargo build -p bashkit-cli --release is reported in the PR description (the cat/tac POC was +9 KB declared-but-unused; ls is fully wired so a real number is meaningful).

Out of scope

• Implementing every uu_ls option (--context, --si, --block-size, full coloration). Leave them parser-accepted-with-explicit-error or wire them to existing bashkit logic only when straightforward.
• Vendoring uucore::quoting-style for --quoting-style semantics — separate question, see the module-vendor mode issue.

Open questions

• Does the codegen tool's translate!() rewriter need to handle Fluent placeables ({ $count }, { $name }) and selectors that appear in uu_ls's about/help strings? If yes, the port-tool gets a Phase-2 enhancement before this PR can land cleanly.
• The codegen output for ls is large (hundreds of args). Measure the cost of try_get_matches_from on every ls invocation against the current handwritten parser. Cache ls_command() in a OnceLock if the construction overhead is meaningful.

Related

• feat(builtins): port uutils argument surfaces via codegen (POC: cat, tac) #1529 — codegen POC (cat, tac)
• Differential testing harness for ported coreutils builtins (vs. uutils binaries) #1530 — differential testing harness (will gain an ls row once this lands)
• specs/coreutils-args-port.md
• crates/bashkit/src/builtins/ls.rs
• crates/bashkit-coreutils-port/