test(session): remove platform and AsyncMock flakes

[bolichen97]

Problem / Motivation

The session test file had three independent sources of platform or teardown noise:

• one security test assumed the host could create symlinks, which fails on Windows without SeCreateSymbolicLinkPrivilege;
• provider factories used bare AsyncMock objects for synchronous methods and for an async-iterator method, leaking unawaited coroutine warnings under strict warning handling;
• one background-session test started a real SessionManager cleanup task without closing the manager.

These are test-contract defects, not reasons to retry or weaken CI.

What changed

• Apply the repository's existing requires_symlinks capability probe to the exact symlink test. The probe actually attempts symlink creation; it does not guess from the OS name or catch arbitrary failures.
• Give the provider doubles synchronous implementations for runtime_info, context_window_tokens, and is_process_alive.
• Return an empty async generator from stream_command, matching the production async-iterator protocol.
• Make the local warm-pool and replacement-provider liveness contracts explicit.
• Close the SessionManager created by test_background_session_name.

The diff is test-only: one file, 17 added lines, one commit.

Determinism

No sleeps, retries, reruns, enlarged timeouts, catch-all skips, or warning filters were added. The symlink skip is capability-based and limited to the one test whose subject requires a symlink.

Validation

• test/test_session.py under strict RuntimeWarning and PytestUnraisableExceptionWarning handling: 290 passed, 2 skipped;
• exact symlink node on this Windows host: deterministic skip with the SeCreateSymbolicLinkPrivilege reason;
• no RuntimeWarning or PytestUnraisableExceptionWarning remains;
• isort, flake8, diff-check, and the official Black changed-file gate: pass;
• merge-tree against main at 901ef09: clean.

The only emitted warning is pytest-xdist's own deprecation notice for the repository's looponfail option; this change does not filter it.

Open-PR overlap

All 238 open PRs were scanned. Six also touch test/test_session.py (#6853, #6600, #6595, #5395, #4550, and #3753), but their patches do not modify these marker, mock-contract, or teardown lines. Pairwise merge checks keep this test file conflict-free; their existing conflicts are in README or production session.py.

Checklist

• One Conventional Commits commit
• Test-only, deterministic scope
• Strict warning validation passes
• No behavior or assertion was relaxed
• No secrets, credentials, or internal references

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed b0283669a198b61963a9ab8d1f029509ae9aed6f — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] b028366

Verdict parsed from the review's SHA-scoped output markers for commit b0283669a198b61963a9ab8d1f029509ae9aed6f.

False positive or not applicable? A repository writer can comment:
/ai-review override fable b0283669a198b61963a9ab8d1f029509ae9aed6f: <one-sentence reason>

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of b0283669a198b61963a9ab8d1f029509ae9aed6f and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] b028366

False positive or not applicable? A repository writer can comment:
/ai-review override gpt b0283669a198b61963a9ab8d1f029509ae9aed6f: <one-sentence reason>

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of b0283669a198b61963a9ab8d1f029509ae9aed6f — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

This is a 17-line test-only flake fix. I've verified the claims: requires_symlinks is the repo's established capability probe (test/conftest.py:95, already used across a dozen test files via the same from conftest import pattern), the sync-lambda mock contract matches the existing convention already present in these factories (is_process_alive = lambda: True with its explanatory comment predates this PR), and the close_all() addition fixes the one test that started a real cleanup task without teardown. The fixes target root causes (AsyncMock returning truthy coroutines for sync/iterator methods; missing teardown; a genuine host-capability dependency) rather than papering over symptoms with reruns or filters, which is exactly what the repo's testing conventions demand. The description matches the diff bidirectionally.

Design-Verdict: PASS

Root-cause flake fixes using the repo's existing capability-probe and mock-contract patterns; no design-level concerns.

[DESIGN-REVIEWED] b028366

[chenmingwei23]

Tier 1 auto-approve: test (1 file): remove platform and AsyncMock flakes in session tests. Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, SAST (Semgrep + CodeQL) annotations/alerts clean, security checklist all-NO, AI reviewers green.

[iamwhatever]

Tier 1 auto-approve: test (1 file). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, SAST annotations clean, security checklist all-NO, AI reviewers green. Category: test-only flake removal in test/test_session.py -- adds missing mock provider attributes (context_window_tokens, runtime_info, stream_command), an is_process_alive stub, close_all() teardown, and a requires_symlinks marker; no production file touched.