Investigate flaky session-manager integration tests (formerly skipIf-CI) #905

Context

packages/codev/src/terminal/__tests__/session-manager.test.ts has 3 integration tests that were originally guarded by `it.skipIf(!!process.env.CI)` — meaning they were skipped on CI but expected to run locally. They legitimately spawn real shellper processes.

While working on #903, the builder observed these 3 tests timing out locally (15s timeout exceeded) and proposed converting them to plain `it.skip`. That change was reverted from PR #904 as out-of-scope. This issue tracks proper investigation.

Affected tests

In `packages/codev/src/terminal/tests/session-manager.test.ts` (line numbers as of revert):

1. `'respects maxRestarts limit'` (around line 1020)
2. `'logs session exit without stderr tail (stderr goes to file)'` (around line 1724)
3. `'no stderr tail logged for file-based stderr (Bugfix Shellper processes do not survive Tower restart despite detached:true #324)'` (around line 1834)

What to investigate

• Reproduce the 15s timeout locally. Is it environment-specific (WSL, slow disk, node-pty build) or universal?
• Is the timeout legitimate (test needs more time) or pathological (deadlock, missing event, async leak)?
• Should these tests:
  • Get a longer timeout and stay enabled locally, or
  • Move to a separate `pnpm test:integration` script and be excluded from default `pnpm test`, or
  • Be rewritten to not spawn real processes, or
  • Be deleted if redundant with other coverage
• Original author's intent: why was the original `skipIf(CI)` chosen? Check `git blame` for context.

Out of scope

• Don't simply re-skip and call it done — the point of this issue is to actually understand and decide.

Triggered by

• Bug porch done is not idempotent on 'verified' state — writes redundant status.yaml + commit #903 / PR fix(porch): make done idempotent on verified terminal state #904 review surfaced this; builder's well-intentioned skip was reverted to keep PR scope clean.