Give input/ and main_helpers/ real test coverage and their own coverage gate

[stphung]

Background

From the testing strategy review (docs/reviews/testing-strategy-review.md): src/input/ and src/main_helpers/ (plus src/tick_events.rs) are mod-declared only in src/main.rs, never in src/lib.rs. That means the ~3,000 integration tests under tests/ — which link against the quest library crate — structurally cannot reach these modules, no matter what coverage invocation CI uses. Their only possible coverage source is inline #[cfg(test)] blocks inside the binary crate itself, and right now that's thin to nonexistent for the highest-risk files.

Measured directly with cargo-llvm-cov (full run, no exclusions):

File	Line coverage
main_helpers/input_routing.rs	0.00%
main_helpers/persistence.rs	0.00%
input/soulforge_input.rs	0.00%
input/stormglass_input.rs	0.00%
input/minigame_input.rs	0.00%
input/haven_input.rs	5.68%
input/deep_input.rs	27.11%
input/mod.rs	28.47%
input/loom_input.rs	53.85%
input/voyage_input.rs	57.08%
input/prestige_input.rs	71.92%
input/harness.rs	85.35%
input/time_vault_input.rs	95.30%

time_vault_input.rs and harness.rs prove the pattern works when applied — the input-replay harness (src/input/harness.rs, extended in replay_tests.rs) can drive these handlers and assert on the resulting InputResult/state. It just hasn't been extended to the 0%/near-0% files yet.

Why this matters

main_helpers/input_routing.rs is the layer that turns an InputResult into an actual save/git-commit (route_game_input(), dispatch_time_vault_action(), save_and_commit, reload_game_state, refresh_vault_browser). A wrong branch here is silent save/commit loss with no automated tripwire — the input-replay harness currently asserts the InputResult variant but stops one layer short of asserting the routing actually performs the save.

Proposed work

2. Write tests for the confirmed gaps, prioritized by risk:

• main_helpers/input_routing.rs — assert route_game_input()/dispatch_time_vault_action() actually perform save/commit/reload for each InputResult variant that should trigger one (highest priority — silent-data-loss risk)
• input/soulforge_input.rs
• input/stormglass_input.rs (also fix the direct SystemTime::now() call flagged in the review while touching this file)
• input/minigame_input.rs
• main_helpers/persistence.rs
• input/haven_input.rs

3. Once (2) has real content, give the binary crate its own coverage gate so input//main_helpers//tick_events stop being invisible to CI:

• Add a second cargo-llvm-cov invocation scoped to the bin target (or move stable, testable logic into lib.rs-reachable modules where that's a more natural fit than a parallel gate)
• Wire it into both ci.yml's coverage job and scripts/ci-checks.sh, keeping the two in sync
• Set an initial threshold based on where coverage actually lands after (2), not a number picked in advance

Do not attempt (3) before (2) — gating on it while these files are still near-0% would just fail the build for no benefit.