fix(shared-writer): make blocker/label/relation mutations idempotent (#600)

[dollspace-gay]

Summary

Fixes GH#600. SharedWriter::{add,remove}_{blocker,label,relation} now short-circuit before write_commit_push when no change is needed, so crosslink issue block <a> <b> exits 0 when <a> is already blocked by <b> (previously failed with a confusing git commit […] in cache failed: error caused by the empty index path — see #601 for the underlying stdout-capture bug).

Changes

• Bug fix (shared_writer/mutations.rs): All six mutations gain an early-return guard that loads the issue, checks for the no-op condition, and returns without touching git. Signatures change Result<()> → Result<bool> (true = mutated, false = no-op short-circuit).
• UX consistency (commands/{deps,label,relate}.rs): Writer branches now use the new bool to print differentiated messages (Dependency already exists, Label 'X' already exists on issue …, Issues X and Y are already related, plus symmetric remove-not-present messages) — matching what the non-writer branches already did.
• Tests (shared_writer/tests.rs): Tightened test_add_label_idempotent (it was using let _ = to swallow the bug, with a // duplicate -- may error on empty commit comment). Added five new regression tests: test_remove_label_idempotent, test_{add,remove}_blocker_idempotent, test_{add,remove}_relation_idempotent.
• Pre-existing clippy cleanup (5 unrelated files): Fixed 20 clippy::pedantic warnings that were already on develop, so cargo clippy --lib --bin crosslink --tests -- -D warnings passes again. lock_check.rs had 11 instances of the same map(|s| !s.success()).unwrap_or(true) pattern; others touch container.rs, compaction.rs, sentinel/collect.rs, sync/tests.rs.

Why Result<bool> instead of silent short-circuit

The reported defect is about exit code, but the writer-branch UX was also inconsistent with the non-writer branch (which already returns bool via db.add_dependency etc. and prints Dependency already exists accordingly). Threading the bool through gives the writer-branch the same honest feedback rather than silently printing "Issue X is now blocked by Y" on a no-op.

Out of scope

• git_commit_in_cache_with_args reports empty error messages for the most common failure mode #601 (stdout dropped from git_commit_in_cache_with_args error messages) — fixing this would partially close crosslink issue block exits non-zero when the blocker is already set #600 symptoms but is its own bug.
• crosslink integrity hydration --repair deletes SQLite rows when the JSON files do not have them #602, add_blocker flow is not transactional across JSON / git / SQLite #604 (hydration / transactionality) — listed as related in crosslink issue block exits non-zero when the blocker is already set #600 but separate concerns.

Test plan

• cargo test --lib — 1888 passed
• cargo test --bin crosslink — 2864 passed (includes the new 6 idempotency tests and the existing test_block_duplicate / test_add_duplicate_label / test_add_duplicate_relation that exercise the non-writer branch)
• cargo clippy --lib --bin crosslink --tests -- -D warnings — clean
• Manual repro from the issue: crosslink issue block L2 L1 twice in a row → both exit 0, second prints Dependency already exists

🤖 Generated with Claude Code