Add mapbox config list/unset alongside get/set

[mattpodwysocki]

Summary

Stacked on #35 (mapbox config get/set) — base branch is feat/106-persisted-config, so this diff is just the incremental change. Rebase onto main once #35 merges.

• mapbox config list: every setting and its current value in one call, rather than one key at a time.
• mapbox config unset <key>: clears a key back to "never set" rather than writing its current default value explicitly. The two read identically through resolve() today, but they're not the same fact on disk — a later default change reaches a cleared key and not one a caller pinned to the old default on purpose.
• get/list now share one resolve() so the two can't answer a key differently; unset gets a clear() counterpart to set's Some(enabled).

Test plan

• cargo build, cargo fmt, cargo clippy --all-targets -- -D warnings, cargo test — all clean
• New unit tests in src/config.rs for resolve/clear
• New integration tests in tests/config.rs: list reports the one known key at its default and after a set, in both text and JSON; unset is proven to remove the key from the file rather than just re-writing its default value (checked against the raw file contents, not just the CLI's own read-back)
• docs/commands.md and CHANGELOG updated

[zmofei]

Small, well-tested PR overall — the write/persist logic, schema/completion generalization, and new tests all check out, and I didn't find any correctness bugs.

One thing worth fixing before merge:

• resolve(), clear(), and set() each have their own match key { UPDATE_CHECK_KEY => ..., _ => unreachable!(...) } block. Adding a second config key now means updating three separate match sites (plus KEYS), and missing one still compiles — it just panics at runtime instead. A single per-key table (key, getter, setter) would make adding a setting a one-line change and remove that footgun.

Two minor nits, non-blocking:

• tests/config.rs's module doc still says it covers get/set only, but this PR adds list/unset tests to the same file.
• list() calls resolve() twice per key (once for entries, once for text); a single pass would avoid the duplicate work as more keys are added.