[codex] optimize sqlite bind fast path

[mattn]

This PR now includes two hot-path reductions.

The first change avoids rebuilding driver.NamedValue slices and unnecessary named-parameter lookup work on the common positional-argument path.

The second change reduces cgo traffic in Rows.Next by fetching one row's column metadata and raw values in a single C helper call before converting them on the Go side.

These results were measured on the same machine.

benchmark	before	after	delta
BenchmarkExec	1893 ns/op	1674 ns/op	-11.6%
BenchmarkQuery	4356 ns/op	3992 ns/op	-8.4%
BenchmarkParams	4845 ns/op	4359 ns/op	-10.0%
BenchmarkStmt	2911 ns/op	2878 ns/op	-1.1%
BenchmarkRows	143348 ns/op	135534 ns/op	-5.5%
BenchmarkStmtRows	137675 ns/op	131549 ns/op	-4.5%
BenchmarkQueryParallel	1809 ns/op	1811 ns/op	+0.1%

benchmark	before allocs	after allocs	before bytes	after bytes
BenchmarkParams	26	24	1104 B/op	848 B/op
BenchmarkStmt	24	23	920 B/op	824 B/op
BenchmarkExec	6	6	128 B/op	128 B/op
BenchmarkQuery	22	22	688 B/op	688 B/op
BenchmarkRows	524	524	11128 B/op	11128 B/op
BenchmarkStmtRows	523	523	11104 B/op	11104 B/op
BenchmarkQueryParallel	16	16	592 B/op	592 B/op

Validated with go test ./..., go test -run '^$' -bench . -benchmem, and go test -run '^$' -bench 'BenchmarkSuite/(BenchmarkRows|BenchmarkStmtRows)$' -benchmem.