Revert #106 !
I think this change was a misunderstanding : the test data does not show clearly that these were the wrong numbers -- but other (real) data since encountered does.

( N.B. the actual Grib spec is really vague on this, hence the interpretation depended on what I found in actual files ! )