...ementation
Because we use `encoding/csv.Reader.ReuseRecord`, each call of `csv.Reader.Read()` return value `[]string`
is a cached slice. Given we need to potentially cache multiple lines thus multiple calls to `Read()`,
what's in the `linesBuf.record` is always duplicate!!
We could fix this problem trivially by turning `ReuseRecord` off, but that would incur an allocation
cost for vast majority of single-line csv operation. That is completely undesired.
So instead, we ourselves cache all the returned strings from (potentially multiple) calls to
`csv.Reader.Rdad()` into `reader.records []string` slice, that managing that buffer ourselves, thus
practically eliminate the over-allocation problem. Accordingly, in the `reader.linesBuf`, instead of
having a `record []string`, we have `recordStart` and `recordNum` to reference into the `reader.records`.