feat(devex): raise the quality bar for AI-assisted contributions

Motivation

RTK contributions are mainly delegated to AI coding agents (Claude Code, Cursor, Copilot, etc.). In that context, the quality of the Rust codebase matters more than usual — the agent can't visually review its output, it relies entirely on static analysis to catch issues.

The current lint pipeline (cargo clippy --all-targets -- -D warnings) only covers Clippy's default group. Three high-value groups are completely invisible today, and two common supply chain checks are absent. This means an AI agent generating a PR can introduce idiomatic Rust violations, unused deps, broken doc links, or even a CVE — and nothing in CI will catch it.

Proposed changes

1. clippy::pedantic + clippy::nursery

Enable the two opt-in Clippy groups that catch idiomatic Rust violations:

``` cargo clippy --all-targets -- -D warnings -W clippy::pedantic -W clippy::nursery ```

The first run surfaces ~2,100 warnings. Most are mechanical fixes (format string inlining, unnecessary raw string hashes, redundant closures, single-char patterns) applied automatically.

Lints that are legitimately inapplicable to a CLI proxy are suppressed in a single #![allow(...)] block at the top of src/main.rs, each with a comment explaining why:

Category	Lints	Reason
Cast lints	cast_precision_loss, cast_possible_truncation, cast_sign_loss, cast_possible_wrap	Intentional usize→f64 for percentage calculations and i64→usize for SQLite results — controlled, small-value casts
Doc lints	doc_markdown, missing_errors_doc, missing_panics_doc	RTK is a CLI tool, not a library — library-style doc requirements don't apply
Naming	module_name_repetitions	git::GitArgs is the natural naming for CLI arg structs
Style	option_if_let_else, cognitive_complexity, too_many_lines, missing_const_for_fn, suboptimal_flops, float_cmp	Reduce readability in filter parsing code; missing_const_for_fn is a nursery lint with many false positives
Signatures	must_use_candidate, needless_pass_by_value, unnecessary_wraps, fn_params_excessive_bools, trivially_copy_pass_by_ref, return_self_not_must_use	Cosmetic on 50+ run() entry points; not worth changing across all modules
Established patterns	format_push_string, non_std_lazy_statics	263 push_str(&format!(...)) and 23 lazy_static! usages — migrating these is tracked as follow-up (see below)
Minor style	use_self, manual_let_else, items_after_statements, similar_names, map_unwrap_or, or_fun_call, match_same_arms, branches_sharing_code, struct_field_names, comparison_chain, redundant_pub_crate, used_underscore_binding	Idiomatic alternatives are less readable in RTK's filter pattern code

Open to discussing any of these before landing — especially the "established patterns" row, which represents real technical debt rather than genuine false positives.

2. cargo-deny — supply chain security

Add a deny.toml covering:

• Advisories: deny known CVEs and yanked crates
• Licenses: explicit allowlist (MIT, Apache-2.0, BSD-*, ISC, MPL-2.0, Zlib…)
• Bans: deny tokio, async-std, futures — enforces RTK's no-async rule at tooling level, not just documentation
• Sources: crates.io only, deny unknown registries and git deps

3. cargo-machete — unused dependencies

Detects unused entries in Cargo.toml. On RTK's current dep tree nothing is removed (all 21 deps are in use), but the check catches future drift.

4. rustdoc -D warnings

``` RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --all-features ```

Catches broken intra-doc links and malformed doc comments. One issue found in the current codebase (<path> in a doc comment interpreted as an HTML tag).

Integration

All four checks wired into CI and a local task:

``` mise run qa → fmt:check + lint + test + doc + machete + deny ```

Questions for discussion

• Crate-level #![allow(...)]: is the table above acceptable as-is, or do you want to address some of these before landing?
• multiple-versions = "warn" vs "deny": the windows-sys family ships 4 simultaneous versions via transitive deps (dirs, ring, colored, clap). Worth trying to reduce, or skip and warn?
• clippy::nursery: some nursery lints are experimental and occasionally noisy. Open to dropping it and sticking with just pedantic if you'd prefer a more stable baseline.

Out of scope for this PR

• lazy_static! → std::sync::LazyLock migration (23 usages — separate modernization task)
• push_str(&format!(...)) → write!() migration (263 usages — separate task)
• Fixing too_many_arguments on run() entry points (requires API design decisions)

A working implementation is available on branch devex/alignment-claude-hook.