test: retry the Keychain store-wide reads that raced their own writes

[StuartMeeks]

Fixes the red main.

What happened

The post-merge run of #78 failed on test (macos-14):

failed KeychainCredentialManagerTests.GetProviderNamesAsync_ReturnsDistinctProviders
  Assert.Equal() Failure: Values differ
  Expected: 2
  Actual:   0

The same commit passed that same leg on the PR run minutes earlier, so it is timing rather than a regression. It is also not a consequence of the #55 owner-attribute change: when no owner is recorded IsOwnedByThisApp falls back to the prefix test, so filtering cannot zero the result.

Root cause — the same mismatch as #61

AddCredentialAsync confirms its write is visible via QuerySingleItem, an exact service+account query. ListCredentialsAsync and GetProviderNamesAsync go through QueryItems/QueryAllItemsForApp, a class-wide query filtered in memory. Visibility to one does not imply visibility to the other, so a read on the second path can legitimately see nothing straight after a write confirmed on the first.

That is precisely the mismatch #61 fixed for the libsecret Restore test. Same shape, different file.

The fix, and why six tests rather than one

Fourteen reads in KeychainCredentialManagerTests already used RetryHelper. Six did not — GetProviderNamesAsync was simply the one that lost the race first, and the other five were equally exposed. All six now retry.

Each predicate matches what the test then asserts, not merely that something appeared:

Test	Predicate
ListCredentialsAsync_ReturnsAddedCredential	r.Any(c => c.AccountId == accountId)
ListCredentialsAsync_FiltersByProvider	r.Count() == 1 per provider
GetCredentialByIdAsync_DoesNotMutateSelection	r.Count() == 2 && r.Any(c => c.IsSelected)
DeleteCredentialAsync_RemovesCredential	!r.Any()
GetProviderNamesAsync_ReturnsDistinctProviders	r.Count() == 2
ListCredentialsAsync_IncludesDisplayFields_FromSummaryProvider	r.Count() == 1

A count-only predicate guarding a content assertion is exactly what let #61 through, so the third row waits on the selection being visible too, not just on two rows existing.

Verification

Build clean, 416 tests locally. The macOS legs are the ones that matter here and only CI runs them — worth watching test (macos-14) and test (macos-15) specifically on this PR, and on main after merge, since a timing fix cannot be proven by a single green run.

🤖 Generated with Claude Code