Add confirm_lsn to replay a load-test backlog from the start

[lukashes]

Problem

Load tests want to generate a backlog once and then measure the reader against the same data repeatedly. Today the first make start-outboxx consumes the backlog and confirms the LSN, so the slot advances and the WAL is recycled; a second run reads nothing.

Solution

• Add confirm_lsn to [source.postgres], default true. When false, sendFeedback keeps sending (so the keepalive and liveness path are unchanged) but always confirms the first LSN seen, so the slot never advances and PostgreSQL keeps the WAL. Every restart then re-reads the same backlog from the start.
• The source captures the first change LSN and reports it while pinned; main wires the flag and warns at startup when it is off (the mode is not durable and must not ship to production).
• The load stand's outboxx config sets confirm_lsn = false, and its image builds with -Dlog_level=info to keep the run quiet.

Workflow: make load once, then make start-outboxx as many times as you like.

Verification

New integration test drives a real slot: with confirm_lsn = false, after reading a multi-row batch and calling sendFeedback(last_lsn), the slot's confirmed_flush_lsn settles on the first LSN, below the batch's last.

[github-actions]

📊 Benchmark Results

Current run is the minimum over 3 passes, compared against the base branch (main @ 10c320b), built on the same runner.

Benchmark	Baseline	Current	Δ Time	Allocs	Status
Converter DELETE	132.21μs	130.88μs	-1.0%	14 → 14	➡️
Converter INSERT	126.92μs	125.91μs	-0.8%	14 → 14	➡️
Converter UPDATE	178.38μs	176.89μs	-0.8%	23 → 23	➡️
JsonSerializer	42.09μs	41.55μs	-1.3%	3 → 3	➡️
KafkaProducer produce	206.92μs	203.86μs	-1.5%	1 → 1	➡️
KafkaProducer sendMessage	248.80μs	245.58μs	-1.3%	0 → 0	➡️
PgOutputDecoder	95.23μs	94.94μs	-0.3%	6 → 6	➡️
getPartitionKeyValue boolean	16.74μs	16.52μs	-1.3%	1 → 1	➡️
getPartitionKeyValue integer	51.29μs	51.11μs	-0.3%	3 → 3	➡️
getPartitionKeyValue not found	0.03μs	0.03μs	+0.0%	0 → 0	⚪ noise
getPartitionKeyValue string	18.57μs	18.46μs	-0.6%	1 → 1	➡️
matchStreams found	17.81μs	17.60μs	-1.1%	1 → 1	➡️
matchStreams not found	0.06μs	0.06μs	+0.0%	0 → 0	⚪ noise

Summary: ➡️ 11 neutral · ⚪ 2 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.}