Problem

Windows/Linux CI intermittently reddens on t.TempDir auto-cleanup even though every assertion in the test passed:

• Windows: TempDir RemoveAll cleanup: ... The process cannot access the file because it is being used by another process.
• Linux: TempDir RemoveAll cleanup: ... directory not empty

It hits different tests each run — internal/boot legacy-migration tests and desktop capabilities tests — which is why reruns kept landing on it (the recent [1/6]..[6/6] series got blocked on it repeatedly).

#3371 attributed this to t.Chdir and switched a single test to WorkspaceRoot, but that same test (TestBuildMigratesLegacySessionsFromConfigSessionDir) kept failing afterward — so the real cause is a teardown race, not the process cwd.

Root cause

The failing tests all build a full Controller. At teardown a background resource can still hold a file under the temp dir for a few milliseconds after Close() returns, and t.TempDir's automatic RemoveAll races it.

Fixes (two, independent)

1. jobs.Manager.Close is now graceful. It previously only cancel()-ed the session context and returned without waiting for job goroutines to observe the cancel and exit. Added a sync.WaitGroup: Start does Add(1)/defer Done(), Close does cancel() then Wait(). Jobs observe the cancel through their run context (exec.CommandContext kills a bash job's process), so the wait is bounded.

2. robustTempDir test helper (boot + desktop) that retries RemoveAll for a short window, absorbing the brief post-Close interval where a background resource still holds a file. A dir that never frees is t.Logf-ed, not fatal, so a genuine leak stays visible without reintroducing the flake.

Verification

• gofmt clean; go vet clean.
• internal/jobs, internal/control, internal/boot pass; the two previously-flaky boot tests pass -count=200 locally.
• desktop test binary compiles.
• Real validation is this PR's own Windows + desktop CI (the flake only reproduces there).

Follow-up

Unblocks #3367 and #3370 (the last two of the [1/6]..[6/6] series), which were red purely on this flake. Once this lands, those two get rebased onto it and should go green normally.