test: retire long-tail cases whose assertion is decided by the test itself (#24132)

Resumes the backlog sweep at a chunk size that actually gets read. Six auditors, 84 files
each, and all six read their full scope case-by-case against production — the first wave
where every chunk closed with no gap. 33 case declarations removed across 22 files, 1 test
file deleted, 826 lines gone. No production code touched.

This wave exists because a conclusion of mine was wrong. I had recorded that yield collapsed
~36x and that deletion was no longer the high-value work. I was dividing cases removed by
files IN SCOPE while the fraction auditors actually READ fell from 100% to about 4%, because
I kept handing them 300-800 files. Recomputed against files read, yield has been flat at 4-7
per 100 with no downward trend. This wave came in at 8.2.

The most instructive removal looked like the most valuable test in scope.
`orchestration-worker-release-reap-fixed.func.test.ts` cites a production bug by two
identifiers, describes orphaned PTYs accumulating until `TasksMax=4096` aborts processes on
EAGAIN, and advertises itself as the functional tier wiring the real orchestration RPC
surface, the real `OrchestrationDb` and the real release modules. Deleting it leaves no
reference to that bug anywhere in `src`.

It still had to go: its fake runtime performed the fence it asserted —

    if (pty.incarnationId !== inc) { return null }
    handleTable.set('term_reminted', { ptyId, epoch: rendererGraphEpoch })

— so the case checking that a reused ptyId with a mismatched incarnation does not resolve was
checking a decision its own spy made twenty lines earlier. The real fence is owned by
`orca-runtime-terminal-handle-incarnation.test.ts:257`, and the other two cases replay
`orchestration-worker-release-incarnation-fallback.test.ts` (which uses a plain
`mockReturnValue` rather than reimplementing the remint) and `worker/worker-release.test.ts:23`.
"Integration test" and "wires real modules" describe the scaffolding, not the asserted step.

Other removals: a self-comparison disguised by an alias, where
`export const getIssueOwnerRepo = getOwnerRepo` makes a case asserting the two "agree" into
`f(x) === f(x)`; four cases whose `vi.mock` of `resolveIssueSource` made both the preference
value and the topology inert; five verdict-precedence cases owned by a verdict-agnostic block;
three call-shape probes on one-line store pass-throughs whose real contracts are driven by
behavioural neighbours; and a `export type _Ref = [...]` declaration whose own comment admits
it exists only to preserve test-only module-surface references.

Kept after checking production rather than shape. An auditor found two near-identical
ten-reconnect loops and kept both: one uses a test-local live-lease filter, the other the
shipped `sshRemotePtyLeaseAllowsReattach` predicate, and the file's own comment explains the
duality is deliberate "so the two cannot drift". Another kept a paths-alignment case that
looks like a validator tested against its own list, because adding a generated file without
registering its path does fail it — and `shellReadyWrappersExist` uses that registered list to
decide whether a partial tree needs regeneration.

Production duplication is now confirmed four times over, and it is why mirrored tests exist:
`createUpdateWorktreeLineage`/`createAssignWorktreeParent` differ by one `console.error`
string; `terminal-path-tap.ts` and `document/path-tap.ts` carry hand-maintained copies of
`matchFilePathAtColumn` under a docblock reading "keep the two in sync". In those cases both
test sides are load-bearing and the duplication belongs on a refactor list.

`mobile/tests-typecheck-baseline.txt` loses one entry. Trimming
`relay-host-signed-out-verdict.test.ts` made it typecheck clean, so the ratchet required
pruning its grandfathered entry — the file graduates from exempt to enforced. Baseline is now
124 entries, down from 125.

Verified: 690 test files / 7,560 cases pass across the touched desktop areas; the modified
mobile files pass (162 cases); `check-tests-typecheck-ratchet.mjs` OK (898 files in program,
124 grandfathered); `check-reliability-gates.mjs` 140 gates; the deleted file is absent from
the gate manifest, `cloud/package.json` and the mobile baseline; nothing under
`mobile/src/test-support/rpc-recording/` or `mobile/rpc-foundation/goldens/` touched.

src/main/github/client-issue-source.test.ts

@@ -730,23 +730,6 @@ describe('GitHub issue source split', () => {
       expect(result.issueSourceFellBack).toBeUndefined()
     })
 
-    it("preference='auto' + no upstream → queries origin", async () => {
-      resolveIssueSourceMock.mockResolvedValueOnce({
-        source: { owner: 'solo', repo: 'orca' },
-        fellBack: false
-      })
-      getOwnerRepoMock.mockResolvedValueOnce({ owner: 'solo', repo: 'orca' })
-      ghExecFileAsyncMock.mockResolvedValueOnce({ stdout: '[]' }).mockResolvedValueOnce({
-        stdout: '[]'
-      })
-
-      await listWorkItems('/repo-root', 10, undefined, undefined, 'auto')
-
-      expect(ghExecFileAsyncMock).toHaveBeenNthCalledWith(1, issueSearchArgs('solo/orca'), {
-        cwd: '/repo-root'
-      })
-    })
-
     it("preference='auto' + upstream exists → PRs query upstream too", async () => {
       // Why: fork-contribution PRs live on the upstream repo — the fork's own
       // PR list is almost always empty. 'auto' must resolve PRs upstream-first
@@ -799,22 +782,6 @@ describe('GitHub issue source split', () => {
       )
     })
 
-    it("preference='upstream' + upstream exists → queries upstream", async () => {
-      resolveIssueSourceMock.mockResolvedValueOnce({
-        source: { owner: 'stablyai', repo: 'orca' },
-        fellBack: false
-      })
-      getOwnerRepoMock.mockResolvedValueOnce({ owner: 'fork', repo: 'orca' })
-      ghExecFileAsyncMock.mockResolvedValueOnce({ stdout: '[]' }).mockResolvedValueOnce({
-        stdout: '[]'
-      })
-
-      const result = await listWorkItems('/repo-root', 10, undefined, undefined, 'upstream')
-
-      expect(decodedIssueSearchPath(0)).toContain('q=repo:stablyai/orca is:issue is:open')
-      expect(result.issueSourceFellBack).toBeUndefined()
-    })
-
     it("preference='upstream' + no upstream → falls back to origin with fellBack=true", async () => {
       resolveIssueSourceMock.mockResolvedValueOnce({
         source: { owner: 'solo', repo: 'orca' },
@@ -831,36 +798,6 @@ describe('GitHub issue source split', () => {
       expect(result.issueSourceFellBack).toBe(true)
     })
 
-    it("preference='origin' + upstream exists → queries origin (not upstream)", async () => {
-      resolveIssueSourceMock.mockResolvedValueOnce({
-        source: { owner: 'fork', repo: 'orca' },
-        fellBack: false
-      })
-      getOwnerRepoMock.mockResolvedValueOnce({ owner: 'fork', repo: 'orca' })
-      ghExecFileAsyncMock.mockResolvedValueOnce({ stdout: '[]' }).mockResolvedValueOnce({
-        stdout: '[]'
-      })
-
-      await listWorkItems('/repo-root', 10, undefined, undefined, 'origin')
-
-      expect(decodedIssueSearchPath(0)).toContain('q=repo:fork/orca is:issue is:open')
-    })
-
-    it("preference='origin' + no upstream → queries origin", async () => {
-      resolveIssueSourceMock.mockResolvedValueOnce({
-        source: { owner: 'solo', repo: 'orca' },
-        fellBack: false
-      })
-      getOwnerRepoMock.mockResolvedValueOnce({ owner: 'solo', repo: 'orca' })
-      ghExecFileAsyncMock.mockResolvedValueOnce({ stdout: '[]' }).mockResolvedValueOnce({
-        stdout: '[]'
-      })
-
-      await listWorkItems('/repo-root', 10, undefined, undefined, 'origin')
-
-      expect(decodedIssueSearchPath(0)).toContain('q=repo:solo/orca is:issue is:open')
-    })
-
     it('surfaces upstreamCandidate in sources regardless of effective preference', async () => {
       // Why: the renderer selector needs to keep rendering after the user picks
       // 'origin'. That requires the envelope to carry the raw upstream even