GTST-19 refactor: improve coding style with pedantic lints and shared helpers

[MonteYin]

Summary

• Add lint/format tooling config (rustfmt.toml, clippy.toml, rust-toolchain.toml, Cargo.toml [lints]) with clippy::pedantic enabled
• Extract shared git2 helpers into git/mod.rs (diff_opts_with_untracked, delta_path, hunk_header, is_binary_delta)
• Add LineTag::from_origin(), PROTOCOL_VERSION constant, and dispatch() helper to reduce duplication
• Deduplicate test setup into tests/common/mod.rs
• Net -290 lines (348 added, 638 removed)
• Bump version to 0.1.2

Test plan

• cargo fmt --check passes
• cargo clippy --all-targets — zero warnings
• cargo test — all 82 tests pass
• cargo build --release — binary runs correctly (--version, diff, stage, status, protocol)
• Manual stage test on real repo verified

[MonteYin]

Review -- Opus 4.6

Review Result

Overall Verdict: PASS

---

1. Correctness

Verdict: PASS

All refactoring changes preserve existing behavior:

• diff_opts_with_untracked() in status.rs: The original code used DiffOptions::new() without explicit context_lines(3), relying on libgit2's default (which is 3). The shared helper now explicitly sets context_lines(3), making the implicit explicit. No behavioral change.
• print_json() helper in output.rs: Both old and new code pass &DiffOutput (a reference) to Response::success(), creating Response<&DiffOutput>. Serde serializes references transparently, so JSON output is identical.
• let _ = writeln!(patch, ...) pattern in reconstruct_patch(): fmt::Write for String is infallible, so discarding the Result is safe and idiomatic.
• is_binary_delta() centralization: This helper is now consistently used across all four foreach callsites (diff.rs delta callback, stage.rs scan_hunk_metadata, collect_hunks, and the hunk-level staging scan loop). This resolves the inconsistency I noted in PR GTST-10 feat: integration tests and binary file fix #7 where only 2 of 4 sites had the check.
• LineTag::from_origin(): Exact same match arms as the inline code it replaces. No behavioral change.
• PROTOCOL_VERSION constant: Type u8 matches the Response.version field type. Values unchanged at 1.

Issues

Minor

• No dedicated unit test for LineTag::from_origin(), but it is exercised transitively through integration tests that verify diff output content. Acceptable for a trivial method.

---

2. Architecture

Verdict: PASS

The extraction of shared helpers into git/mod.rs is well-motivated and well-scoped:

• diff_opts_with_untracked() and diff_opts_tracked_only(): Good separation. The two variants have clearly documented use cases. The CONTEXT_LINES constant centralizes a value that was duplicated across 4 callsites.
• delta_path(), hunk_header(), is_binary_delta(): These were copy-pasted across diff.rs and stage.rs (6+ callsites). Extracting them into the parent module is the right scope -- they're git2-specific helpers shared by submodules.
• dispatch() in protocol.rs: Clean dedup of the Ok -> success / Err -> error pattern that was repeated 3 times.
• tests/common/mod.rs: Correct use of mod common + use common::{...} pattern for sharing test utilities across integration test files. The #![allow(dead_code)] is standard practice here.
• resolve_format visibility change (pub to fn): Good. It was only used within output.rs.

---

3. Security

Verdict: PASS

No security-relevant changes in this PR. The refactoring is purely structural -- no new I/O paths, no changes to input validation, no changes to how user input reaches git operations.

---

4. Performance

Verdict: PASS

No performance impact. Helper functions are trivially inlineable. The use std::fmt::Write import inside reconstruct_patch (moved from top-level) has zero runtime cost.

---

5. Simplicity

Verdict: PASS

Net -290 lines is a meaningful reduction. The changes follow KISS and DRY:

• Eliminated 6+ copy-paste sites for delta_path, hunk_header, is_binary_delta, and LineTag matching.
• Eliminated 3 duplicate setup_repo() + gitsift() + parse_json() definitions across test files.
• Replaced verbose match + Option patterns with let ... else (idiomatic Rust 2024).
• rustfmt.toml with max_width = 100 and use_small_heuristics = "Max" produces denser formatting that reduces vertical noise without sacrificing readability.

Issues

Minor

• clippy.toml thresholds (too-many-arguments-threshold = 7, cognitive-complexity-threshold = 30) are generous. These are reasonable for a codebase of this size, but worth revisiting if the codebase grows. Not blocking.
• CI change from cargo clippy --all-targets -- -D warnings to cargo clippy --all-targets: Default clippy warnings will no longer fail CI (only the three deny-level lints will). The intent is to manage lint policy via Cargo.toml [lints] rather than CI flags, which is a valid architectural choice. Worth being aware that this is a deliberate relaxation of CI strictness for non-configured lints.

---

Summary

Clean refactoring PR that achieves its stated goals: shared helpers eliminate copy-paste across diff.rs and stage.rs, test deduplication removes 3 identical setup_repo implementations, and pedantic lints are configured via Cargo.toml rather than ad-hoc. The is_binary_delta() centralization resolves the inconsistency flagged in PR #7. No correctness, security, or performance concerns. The CI clippy change is a deliberate policy shift (lint config in Cargo.toml over CI flags) that is reasonable for a project enabling pedantic lints.