test(github): tighten owner-repo cache-churn and origin-mock hygiene

Round-2 review follow-ups on the branch's own test edits:
- gh-utils WSL cache-separation test used a non-matching "not found"
  error, so the upstream miss took the transient (uncached) path and the
  "cached hit" call silently re-probed; the git-process count pin had
  also been dropped. Use the real "No such remote" wording so the stable
  negative-cache path is exercised, and restore toHaveBeenCalledTimes(4)
  to guard against git-spawn churn.
- Reset getOriginOwnerRepoMock in client.test.ts beforeEach blocks so a
  stale queued once-value cannot leak between getRepoUpstream tests.

Co-Authored-By: Claude <noreply@anthropic.com>

Co-authored-by: Orca <help@stably.ai>

src/main/github/client.test.ts

@@ -169,6 +169,7 @@ describe('getPRForBranch', () => {
     execFileAsyncMock.mockReset()
     ghExecFileAsyncMock.mockReset()
     getOwnerRepoMock.mockReset()
+    getOriginOwnerRepoMock.mockReset()
     getIssueOwnerRepoMock.mockReset()
     getOwnerRepoForRemoteMock.mockReset()
     resolvePRRepositoryCandidatesMock.mockReset()
@@ -3107,6 +3108,7 @@ describe('updatePRState', () => {
   beforeEach(() => {
     ghExecFileAsyncMock.mockReset()
     getOwnerRepoMock.mockReset()
+    getOriginOwnerRepoMock.mockReset()
     ghRepoExecOptionsMock.mockClear()
     githubRepoContextMock.mockClear()
     acquireMock.mockReset()
@@ -3167,6 +3169,7 @@ describe('GitHub GraphQL rate-limit guard', () => {
     execFileAsyncMock.mockReset()
     ghExecFileAsyncMock.mockReset()
     getOwnerRepoMock.mockReset()
+    getOriginOwnerRepoMock.mockReset()
     getIssueOwnerRepoMock.mockReset()
     getOwnerRepoForRemoteMock.mockReset()
     resolvePRRepositoryCandidatesMock.mockReset()

src/main/github/gh-utils.test.ts

@@ -221,11 +221,15 @@ describe('github owner/repo resolution', () => {
   })
 
   it('keeps local host and local WSL owner/repo cache entries separate for the same path', async () => {
-    // upstream-first: each getOwnerRepo call tries upstream then origin
+    // upstream-first: each getOwnerRepo call tries upstream then origin.
+    // Why: use the real "No such remote" wording so the upstream miss takes the
+    // stable-missing negative-cache path (isStableMissingGitRemoteError). That
+    // lets the third call hit cache with zero probes, which the call-count pin
+    // below asserts — guarding against git-process churn (issue #6381 class).
     gitExecFileAsyncMock
-      .mockRejectedValueOnce(new Error('not found'))
+      .mockRejectedValueOnce(new Error("fatal: No such remote 'upstream'"))
       .mockResolvedValueOnce({ stdout: 'git@github.com:host/orca.git\n' })
-      .mockRejectedValueOnce(new Error('not found'))
+      .mockRejectedValueOnce(new Error("fatal: No such remote 'upstream'"))
       .mockResolvedValueOnce({ stdout: 'git@github.com:wsl/orca.git\n' })
 
     await expect(getOwnerRepo('/repo')).resolves.toEqual({ owner: 'host', repo: 'orca' })
@@ -253,6 +257,8 @@ describe('github owner/repo resolution', () => {
       cwd: '/repo',
       wslDistro: 'Ubuntu'
     })
+    // The third (cached) lookup must not re-probe: 2 remotes x 2 distinct runtimes.
+    expect(gitExecFileAsyncMock).toHaveBeenCalledTimes(4)
   })
 
   it('prunes expired distinct owner/repo cache entries on later lookups', async () => {