Improve the handling of chunk-encoded responses

Improve the handling of chunk-encoded responses

... by buffering the data even if filters are disabled and
properly keeping track of where the various chunks are supposed
to start and end.

Previously Privoxy would merely check the last bytes received
to see if they looked like the last-chunk.

This failed to work if the last-chunk wasn't received in one
read and could also result in actual data being misdetected
as last-chunk.

Should fix: SF support request #1739
Reported by: withoutname