test(db): cover dotted-name helper, schema-inspection helpers, orphan sweep

[juacker]

Summary

Adds 21 additive unit tests to src-tauri/src/db/mod.rs::mod tests, covering helpers the existing 12-test block did not exercise directly:

Helper	New tests
is_legacy_dotted_tool_name (pure)	5 — all prefixes, underscore form rejection, external MCP rejection, empty/dotless, prefix boundary
table_exists, column_exists	3 — present/absent semantics on in-memory SQLite
foreign_key_targets, drop_table_if_exists	4 — referenced-table listing, empty-vec for absent table, no-op on absent table
sweep_orphaned_running_state	5 — running → failed, COALESCE keeps existing error, non-running rows untouched, idempotence, empty table
canonicalize_legacy_tool_names (JSON-blob edge cases)	4 — non-array content skipped, parts without tool_name skipped, only first dot rewritten via replacen, mixed legacy/canonical names in same row

Plus two helpers local to the test module (create_workspace_tasks_for_sweep_test, insert_sweep_task) so the sweep tests pin only the columns the production code reads/writes.

Why

• sweep_orphaned_running_state is what prevents stuck `running` workspace_tasks from haunting users after a crash; if it silently no-ops, those rows linger forever.
• is_legacy_dotted_tool_name decides which historic tool names get rewritten on startup; a wrong prefix list breaks OpenAI-compatible providers on saved sessions.
• table_exists/column_exists/foreign_key_targets/drop_table_if_exists gate every migration; a regression silently breaks "is the legacy table present?" checks.
• canonicalize_legacy_tool_names blob path uses `replacen('.', "_", 1)` — the new tests pin that behavior so a refactor to `replace` would fail loudly. (Reviewer noted: the direct-column SQL path uses `REPLACE` which replaces all dots — this is a pre-existing inconsistency on dot-rich names like `bash.exec.advanced`. Out of scope for this PR; tests document the current behavior of each path.)

No production code changes; all additions live inside the existing `#[cfg(test)] mod tests` block. Branch is based on `main` at `7067b08` (Release v26.5.22).

Validation

• `cargo fmt --check` ✅
• `cargo clippy -- -D warnings` (lib-only, CI parity) ✅
• `cargo test --lib` → 344 passing / 1 ignored (was 323/1 before this branch — 22 unit tests gained including helper-driven new variants of `column_exists` checked in one body)
• `npm run lint` → 0 errors (46 unchanged warnings)

Review

Both Code Reviewer members verdict: production_quality (0 blocking findings). Two info-severity notes recorded:

1. Test count framing (the diff adds 21 `#[test]` cases; one "test" in my draft notes was actually a fixture helper). Corrected here.
2. Pre-existing production inconsistency between SQL `REPLACE` (direct column) and Rust `replacen('.', _, 1)` (JSON blob) — out of scope; documented in test names.

🤖 Generated by CLAI QA automation.