fix(config): round-trip the fetchRecurse config key correctly

[bashandbone]

What & why

Fixes the P0-3 finding from the test audit (#62): submod silently dropped the documented fetchRecurse config key and could not read back the fetch-recurse setting it wrote.

Two distinct defects, both real and round-trip-breaking:

1. Read side — the serde field is fetch_recurse with no rename, while the README (and the test suite, and the bundled sample) document the key as fetchRecurse. So a config written per the docs deserialized to None — the value was silently dropped.
2. Write side — both config writers (save_config and write_full_config) emitted fetch = "<to_gitmodules>". That's the wrong key (fetch, neither documented fetchRecurse nor field fetch_recurse) and the wrong value form (to_gitmodules() yields git's true/false, but the deserializer expects the serde always/never). So submod could not parse a config it had just written.

The existing test_change_global_sets_update_and_fetch actually enshrined the bug — it asserted the output contained fetch = "true", with a comment rationalizing it.

Approach

• Standardize on the documented fetchRecurse key via #[serde(rename = "fetchRecurse")] on all three structs (SubmoduleGitOptions, SubmoduleDefaults, SubmoduleEntry). rename (not alias) is required because Config::load merges layers through figment at the raw-dict level — an alias would leave the file layer (fetchRecurse) and the CLI-provider layer (fetch_recurse) as two different keys, which serde then rejects as a duplicate field. A single canonical spelling matches how ignore/update already behave.
• Add SerializableFetchRecurse::as_config_value() returning the serde kebab form (on-demand/always/never), explicitly distinct from to_gitmodules() (git's true/false), and use it in both writers.
• Update the writer key to fetchRecurse and add it to the known-key lists.
• Update the bundled sample_config/submod.toml comments.

Tests (TDD — each watched fail first)

• test_config_toml_fetch_recurse_camelcase_key_is_parsed — deserializes the documented fetchRecurse key and asserts the parsed value (not file text).
• test_write_full_config_fetch_recurse_round_trips — writes config via write_full_config, reloads, asserts Always/Never survive — the core regression test.
• test_fetch_recurse_as_config_value_uses_serde_form_not_git_bools — pins the TOML value form against regressing to the git encoding.
• Corrects test_change_global_sets_update_and_fetch to assert the round-trippable fetchRecurse = "always" and that a subsequent check succeeds.

Full suite: 518 tests pass.

Notes

• config.rs and git_manager.rs also pick up rustfmt (1.8.0) normalization of pre-existing unformatted lines — required to pass the cargo fmt pre-commit gate, which currently fails on ~154 pre-existing diffs across the repo. Reformatting was confined to the two files this PR already touches.
• Follow-ups (not in this PR): the JSON schemas (schemas/v1.0.0, schemas/v1.1.0) still define the key as fetch with additionalProperties: false, so they'd reject the documented fetchRecurse; and schemas/v1.1.0/...json is currently malformed JSON (a stray ""). The repo-wide rustfmt drift (~154 diffs across 13 files) is also worth a standalone style: commit.

Closes the P0-3 item in #62.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 84.09091% with 28 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/config.rs	73.23%	19 Missing ⚠️
src/git_manager.rs	89.88%	9 Missing ⚠️

Files with missing lines	Coverage Δ
src/options.rs	94.76% <100.00%> (+0.11%)	⬆️
src/git_manager.rs	90.36% <89.88%> (-0.17%)	⬇️
src/config.rs	81.97% <73.23%> (+0.25%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.