De-flake TestExecSource on Windows

[ppkarwasz]

TestExecSource.testBatchTimeout fails intermittently on the Windows runner (e.g. this run). Three causes, all Windows-specific or Windows-amplified:

1. Fixed sleeps lose to PowerShell start-up. The tests slept a fixed 1.5–2.5 s before draining the channel; PowerShell cold-start on a loaded runner regularly takes longer. The tests now await the source counter (Awaitility, up to 30 s) via a new package-private ExecSource#getSourceCounter(), following the SyslogTcpSource precedent. This also cuts ~20 s of unconditional sleeping from the suite (~41 s → ~15 s).
2. Write/read race. The input file was written after source.start(), but the Windows command Get-Content <file> | Select-Object -Last 10 is a one-shot read: it sees whatever happens to be in the file when PowerShell gets to it. The file is now fully written before the source starts.
3. Wrong semantics on Windows. Because the one-shot command exits, events were delivered by the EOF flush rather than the 750 ms batch timeout the test is meant to exercise. Get-Content -Tail 10 -Wait is the real tail -f equivalent: the process stays alive and only the timed flush delivers the events.

Also replaces the hard-coded /tmp path, which only worked on Windows by accident, with a real temporary file.