Optimize http writer

[ReverseTM]

This pull request improves the performance of http writer. By changing the data passed to the template, it was possible to reduce the number of allocations by several times. Also a bug was fixed, because of which the error coming from output.Teardown was not processed.

[choseenonee]

Here you are allocating rows slice on each buildRequest call and it defeats much of the benefit of sync.Pool

Consider reusing bodyPayload.Rows 2D slice by getting it from sync.Pool.
Something like that:

// Grab a payload with a ready slice cap.
payload := w.payloadPool.Get().(*bodyPayload)
defer w.payloadPool.Put(payload)

// Reset length to zero, keep cap
payload.Rows = payload.Rows[:0]

// Append each row’s Values slice – no new [][]any heap alloc
for _, dr := range dataRows {
    payload.Rows = append(payload.Rows, dr.Values)
}

[choseenonee]

Also, I think you should defer sync.Pool.Put after Get