Report the shutdown from the batch loop, not the signal handler

[lukashes]

handleShutdownSignal logs from the signal handler. The handler runs on whatever
thread the signal interrupts, so the log rebinds the shared stderr writer under a
call already formatting into it, splitting that line, and it waits on the stderr
lock for as long as a write to stderr takes.

• Keep only the atomic store in the handler.
• Report the shutdown from the batch loop in Processor.run, where the flag is observed.

Closes #83.

[github-actions]

📊 Benchmark Results

Current run is the minimum over 3 passes, compared against the base branch (main @ c445d8a), built on the same runner.

Benchmark	Baseline	Current	Δ Time	Allocs	Status
Converter DELETE	146.11μs	147.60μs	+1.0%	13 → 13	➡️
Converter INSERT	146.14μs	148.40μs	+1.5%	13 → 13	➡️
Converter UPDATE	176.03μs	176.14μs	+0.1%	22 → 22	➡️
JsonSerializer	50.89μs	51.46μs	+1.1%	3 → 3	➡️
KafkaProducer produce	249.10μs	246.86μs	-0.9%	1 → 1	➡️
KafkaProducer send	286.49μs	261.38μs	-8.8%	0 → 0	🟢 faster
PgOutputDecoder	98.43μs	98.97μs	+0.5%	6 → 6	➡️
getPartitionKeyValue boolean	16.23μs	16.23μs	+0.0%	1 → 1	➡️
getPartitionKeyValue integer	0.04μs	0.04μs	+0.0%	0 → 0	⚪ noise
getPartitionKeyValue not found	0.03μs	0.03μs	+0.0%	0 → 0	⚪ noise
getPartitionKeyValue string	17.93μs	18.15μs	+1.2%	1 → 1	➡️
matchStreams found	18.03μs	18.01μs	-0.1%	1 → 1	➡️
matchStreams not found	0.06μs	0.06μs	+1.8%	0 → 0	⚪ noise

Summary: 🟢 1 faster · ➡️ 9 neutral · ⚪ 3 ignored (sub-μs)

_{Thresholds: <1μs ignore · 1–20μs 15% · 20–50μs 10% · ≥50μs 5%. Measured on a shared CI runner — treat small deltas as noise. Informational only; this check never fails the build.}