Deduplicate test fixtures and prune subsumed tests

[robinebers]

TL;DR — Second test-simplification pass after #1143, focused on the suites that pass didn't touch: consolidates duplicated test fixtures, parameterizes near-identical test clusters, and removes tests whose behavior is already verified elsewhere. Net −360 lines (724 deleted, 364 added) with the full suite still green.

What was happening

• Four test files each carried their own private copy of the Cursor JWT builder and an identical key/value SQLite fake; the OpenCode scanner and provider suites duplicated their SQLite stub and row builder; BlockingParser existed twice (integer and Claude-entry variants); and TogglingProviderRuntime / a private SequenceProviderRuntime were the same idea implemented twice.
• Several suites repeated ~25-line provider constructions per test (ClaudeDesktopAuthStore, Antigravity error-category tests).
• A handful of tests were near-identical clones (four Grok Bot optional-endpoint tests, Grok credits absent/non-numeric field pairs, Devin credentials-TOML pair), byte-identical duplicates (a ReorderGeometry test that only renamed a string ID), or restated the implementation in the test loop (two ShareCard condensed-row tests recomputed the production rule instead of asserting expected output).
• A few single tests were fully subsumed by stronger tests elsewhere (PanelHeightCoordinator, AntigravityLayout pins, ZAI empty-limits, GrokCreditsConfig monthly period, MenuBarContent compaction, LocalUsageAPI limits-404, OpenCode zen-only scan, ShellEnvironmentSnapshot roundtrip, ClaudeDesktopAuthStore revoked-CLI fallback pair).

What this changes

• Adds shared makeCursorJWT and KeyValueSQLite to TestSupport.swift; the four Cursor suites now use them.
• Shares the OpenCode SQLite stub (OpenCodeFakeSQLite) and openCodeRow builder between the scanner and provider suites.
• Replaces TogglingProviderRuntime with one shared SequenceProviderRuntime (different snapshot per refresh, call-counting); makes BlockingParser generic over the item type.
• Extracts makeProvider/makeAuthStore helpers in ClaudeDesktopAuthStoreTests and a makeCloudCodeProvider helper in AntigravityProviderTests, collapsing repeated construction boilerplate.
• Parameterizes the four Grok Bot optional-endpoint tests into one table-driven test (all four cases kept), and merges the Grok credits, Devin TOML, and OpenCode hasLocalCredentials pairs.
• Removes duplicated or subsumed tests listed above, each verified as covered elsewhere before deletion; the ShareCard pair is replaced by one fixed-expectation test on a hand-built fixture, and MenuBarContent's unique $130 rounding case moved into MetricFormatterTests where the rule lives.

Heads-up

• Deliberately untouched: LayoutStoreTests and the large Claude/Codex/Copilot suites (every test pins a distinct behavior or documented regression), LogRedactionTests' verbatim Rust-parity cases, all boundary/clamp tests, and every concurrency/ownership/account-isolation test.
• No production code changed — all 25 modified files are under Tests/.

Tests

• swift build --build-tests — clean (only pre-existing warnings).
• swift test — 1,173 tests passed, 0 failures, 3 skipped (pre-existing env-gated live/parity tests), plus the CLI and swift-testing bundles. Baseline run before the changes was also green.

🤖 Generated with Claude Code

---

Note

Low Risk
Only test code under Tests/OpenUsageTests is modified; behavior is unchanged and risk is limited to possible loss of unique test coverage, which the PR explicitly addresses by consolidating rather than deleting scenarios.

Overview
This PR is a test-only cleanup (~−360 lines): no production code changes.

Shared fixtures — makeCursorJWT, KeyValueSQLite, SequenceProviderRuntime (replacing TogglingProviderRuntime), and a generic BlockingParser land in TestSupport.swift / JSONLScannerTestSupport.swift. Cursor suites drop four duplicate JWT/SQLite copies; OpenCode shares openCodeRow and OpenCodeFakeSQLite between scanner and provider tests.

Less boilerplate — makeCloudCodeProvider in Antigravity provider tests and makeAuthStore / makeProvider / cliCredentials in Claude desktop auth tests replace repeated ~25-line setups.

Consolidated cases — Four Grok Bot optional-endpoint tests become one table-driven test; Grok credits absent-field tests, Devin HTTPS-vs-HTTP TOML tests, and OpenCode hasLocalCredentials positive/empty paths are merged. ShareCard condensing is asserted on a fixed hand-built fixture instead of re-implementing the production rule in the test.

Removed redundancy — Tests dropped where coverage already exists (e.g. Antigravity pins exact-set, MenuBar compaction → MetricFormatterTests for $130 tray rounding, duplicate Claude revoked-CLI fallback, LocalUsageAPI limits 404, and others noted in the PR description).

The full suite remains green (1,173 tests).

^{Reviewed by Cursor Bugbot for commit d0771a3. Bugbot is set up for automated code reviews on this repo. Configure here.}