perf(table): reuse RowScanner across Iter rows

[VojtechVitek]

What

Table.Iter created a fresh scany RowScanner on every row via pgxscan.API.ScanRow, recomputing the column-to-field mapping each iteration. This creates one RowScanner before the loop and reuses it, so the reflection work happens once per query instead of once per row.

Motivated by this write-up on pgx.CollectRows — the same per-row reflection anti-pattern it describes was present in Iter. GetAll/List/pagination already reuse a single RowScanner internally, so this only affects the Iter streaming path.

Commits

1. test(table) — extracts the Iter scan loop into iterRows and adds BenchmarkTableIter, driven by an in-memory fake pgx.Rows (no DB needed, runs in CI). Baseline.
2. perf(table) — the one-line-idea change: reuse the RowScanner.

Result

BenchmarkTableIter via benchstat, count=8, all deltas p=0.000:

Metric (1000 rows)	Baseline	Improved	Δ
time	394.8µs	177.5µs	−55%
B/op	406.6 KiB	63.2 KiB	−84%
allocs/op	6002	1008	−83% (~6/row → ~1/row)

Testing

• go test ./tests/... (Postgres-backed) passes, including TestIter and TestTable.
• go build ./... and go vet ./ clean.

🤖 Generated with Claude Code